0 Up votes0 Down votes

10 views71 pagesSep 30, 2016

© © All Rights Reserved

PDF, TXT or read online from Scribd

© All Rights Reserved

10 views

© All Rights Reserved

- Guide Nut 2007 DRAFT 2
- THT 3
- Lec08
- Why Does Higher Q Make Better Clocks_vs2
- Imaging Process Steps
- PPT-7
- BootStrapping.pdf
- Statistik Uji Box M
- Standard Deviation Formulas
- Chapter 4
- norway04_nonparametric
- Rci-10 Scoring and Norms
- MacCallum Et Al 1999 Sample Size in FA
- third revision
- Nie Boer
- 36_1
- 82011987.pdf
- 2-1 a Thomas Homogeneity ISO 13528 Template
- Factor Model Risk Lecture Handout
- Descriptive Statistics

You are on page 1of 71

Petr Adamek

John Y. Campbell

Andrew W. Lo

A. Craig MacKinlay

Luis M. Viceira

Author address:

MIT Sloan School, 50 Memorial Drive, Cambridge, MA 02142{1347

Department of Economics, Harvard University, Littauer Center, Cambridge, MA 02138

MIT Sloan School, 50 Memorial Drive, Cambridge, MA 02142{1347

Wharton School, University of Pennsylvania, 3620 Locust Walk, Philadelphia, PA 19104{6367

Department of Economics, Harvard University, Littauer Center, Cambridge, MA 02138

Contents

List of Figures

List of Tables

Preface

Problems in Chapter 2

Solution 2.1

Solution 2.2

Solution 2.3

Solution 2.4

Solution 2.5

Problems in Chapter 3

Solution 3.1

Solution 3.2

Solution 3.3

Solution 3.4

Solution 3.5

Solution 3.6

Solution 3.7

Problems in Chapter 4

Solution 4.1

Solution 4.2

Solution 4.3

Problems in Chapter 5

Solution 5.1

Solution 5.2

Solution 5.3

Solution 5.4

Problems in Chapter 6

Solution 6.1

Solution 6.2

Solution 6.3

Problems in Chapter 7

Solution 7.1

Solution 7.2

Solution 7.3

Solution 7.4

Solution 7.5

v

vii

1

3

3

3

3

3

4

9

9

10

10

11

13

13

19

23

23

23

24

25

25

26

27

27

29

29

30

30

33

33

34

34

35

36

iii

iv

Problems in Chapter 8

Solution 8.1

Solution 8.2

Solution 8.3

Problems in Chapter 9

Solution 9.1

Solution 9.2

Solution 9.3

Solution 9.4

Problems in Chapter 10

Solution 10.1

Solution 10.2

Problems in Chapter 11

Solution 11.1

Solution 11.2

Problems in Chapter 12

Solution 12.1

Solution 12.2

Solution 12.3

CONTENTS

37

37

39

41

45

45

45

46

48

51

51

52

55

55

58

61

61

62

62

List of Figures

2.1

3.1

3.2

3.3

3.4

3.5

3.6

3.7

Histogram of IBM Price Changes

Histogram of IBM Price Changes Falling on Odd or Even Eighth

Histogram of Times Between Trades for IBM

IBM Price and Volume on Jan 4, 1988

IBM Price and Volume on Jan 5, 1988

IBM Bid/Ask Spread Histogram on Jan 5, 1988

14

14

15

15

17

18

19

12.1

63

vi

LIST OF FIGURES

List of Tables

2.1

2.2

2.3

4

Periods for Daily and Monthly Data

4

Statistics for Daily and Monthly Simple and Continuously Compounded

Returns

5

3.1

3.2

3.3

3.4

3.5

Resulting Statistics for Problem 3.1.3

Simulation Results for Problem 3.5

Unconditional and Conditional Distributions of Bid/Ask Spreads

Contingency Tables

12

12

16

20

21

12.1

62

vii

viii

LIST OF TABLES

Preface

The problems in The Econometrics of Financial Markets have been tested in PhD courses

at Harvard, MIT, Princeton, and Wharton over a number of years. We are grateful to the

students in these courses who served as guinea pigs for early versions of these problems,

and to our teaching assistants who helped to prepare versions of the solutions. We also

thank Leonid Kogan for assistance with some of the more challenging problems in Chapter

9.

PREFACE

Problems in Chapter 2

Solution 2.1

2.1.1 Recall the martingale property given by (2.1.2) and observe that the mean-squared

error of the time-t forecast Xt of price Pt+1 is

2 ; P 2 jPt ; : : : ] :

(S2.1.1)

E[(Xt ; Pt+1 )2 jPt ; : : : ] = (Xt ; Pt )2 + E[Pt+1

t

This expression is minimized by the forecast Xt Pt.

2.1.2 Let l > 0. Then

E[(Pt ; Pt;k )(Pt;k;l ; Pt;2k;l )] = E E[Pt ; Pt;k jPt;k;l ; : : : ]

(S2.1.2)

(Pt;k;l ; Pt;2k;l )

= E[0(Pt;k;l ; Pt;2k;l )] = 0 :

Solution 2.2

(S2.2.1)

RW1 ) RW2 ) M ) RW3;

and no other implication holds in general. For example, consider the following counterexamples. Let fn g1

n=1 be a sequence of random variables drawn independently from a

uniform distribution over the interval [;1; 1] and 0 = 0. Then the process with increments

(i) 2n;1 n and 2n jn j ; 1=2 satises RW3 but not M; (ii) n n n;1 satises M

but not RW2; (iii) n nn satises RW2 but not RW1.

Solution 2.3

A necessary condition for the log-price process pt in (2.2.9) to satisfy RW1 is + = 1. Let

c + and consider the set of all non-RW1 Markov processes (2.2.9), i.e., c 6= 1. The

restriction CJ = 1 is equivalent to = c=4. The constraints 0 ; 1 are satised

exactly for c 2 [1; 4=3] and therefore the set of all two-state Markov chains represented by

the pair (; ) that cannot support any RW1 process but still yields CJ = 1 is simply

p

p

(S2.3.1)

f(1 1 ; c;1 ; 1 1 ; c;1 )c=2; 1 < c 4=3g:

Such Markov chains do generate sequences, reversals, etc.

Solution 2.4

For a stationary process, Var[Zt ] = Var[Zt;k ] and Cov[Zt;k ; Zt;l ] = Cov[Zt ; Zt;l+k ].

Thus, we have

(S2.4.1)

Var[Zt (q)] =

q;1

X

k=0

Var[Zt;k ] + 2

q ;1

X

(q ; k)Cov[Zt ; Zt;k ]

k=1

Cov[Zt ; Zt;k ] are

simply the number of k-th

order autocovariance terms in the variance of the multiperiod return Zt (q) (recall that this

multiperiod return is the sum of q one-period returns). The coecients decline linearly

3

PROBLEMS IN CHAPTER 2

Ten individual stocks used for problem 2.5, identied by CRSP permanent number

PERMNO,

CUSIP identier, (most recent) ticker symbol and abbreviation of full name.

PERMNO

CUSIP

18075

30840

26470

32096

19174

12095

15747

12490

18286

15472

03203710

21161520

29265N10

36480210

37006410

37083810

45870210

45920010

75510310

98252610

Ampco-Pittsburgh Corp.

Continental Materials Corp.

Energen Corp.

Garan Inc.

General Host Corp.

General Signal Corp.

Interlake Corp.

International Business Machs. Corp.

Raytech Corp. De

Wrigley, William Jr. Co.

AP

CUO

EGN

GAN

GH

GSX

IK

IBM

RAY

WWY

Daily Periods

Monthly Periods

A

A1

A2

A3

A4

620703{941230

620703{700923

700924{781027

781030{861128

861201{941230

8179

2045

2045

2044

2045

620731{941230

620731{700831

700930{780929

781031{861031

861128{941230

390

98

97

97

98

with k until they reach zero for k = q because there are successively fewer and fewer

higher-order autocovariances.

From (2.4.19) it is apparent that individual autocorrelation coecients can be nonzero but their weighted average can be zero. For example, according to (2.4.19), VR(3) =

1 + 2( 23 1 + 13 2 ), hence a non-random-walk process with 1 = ; 41 and 2 = 21 will satisfy

VR(3) = 1. Therefore, the variance ratio test will have very lower power against such

alternatives, despite the fact that they violate the random walk hypothesis.

Solution 2.5

We consider the daily and monthly returns of the ten individual stocks considered in

Chapter 1 (see Table 1.1). We use CRSP daily data consisting of 8,179 days from July 3,

1962 to December 30, 1994 and CRSP monthly data consisting of 390 months from July

31, 1962 to December 30, 1994. For these ten stocks there are 23 missing daily returns

and 4 missing monthly returns in our sample. The stocks are identied in Table 2.1, and

we shall refer to them by their ticker symbols (value-weighted and equal-weighted indexes

will be denoted by VW and EW).

Denote the entire sample period by A and the four consecutive subperiods of approximately equal length by A1 , A2 , A3 , A4 , respectively (note that these periods dier for

daily and monthly data). Descriptions of lengths and starting and ending dates of the

periods are given in Table 2.2.

SOLUTION 2.5

Statistics for daily and monthly simple and continuously compounded returns. All the statistics ^, ^ , and ^(1) are reported in percent.

Security

Period

VW

EW

AP

CUO

EGN

GAN

GH

GSX

IK

IBM

RAY

WWY

A

A1

A2

A3

A4

A

A1

A2

A3

A4

A

A1

A2

A3

A4

A

A1

A2

A3

A4

A

A

A21

A

A43

A

A1

A2

A3

A4

A

A1

A2

A

A43

A

A1

A2

A3

A4

A

A

A21

A3

A4

A

A1

A2

A3

A4

A

A1

A2

A

A43

A

A1

A2

A

A43

Simple Returns

Daily Sampling

Monthly Sampling

^

^ ^(1)

^

^ ^(1)

0.044 0.803 19.4 0.96 4.37

4.8

0.035 0.635 25.9 0.76 3.82

6.2

0.026 0.814 29.4 0.71 4.65

5.3

0.070 0.821 16.0 1.40 4.58 -4.7

0.045 0.913 11.0 0.95 4.36

9.8

0.078 0.685 38.6 1.25 5.67 22.0

0.069 0.728 38.8 1.18 5.44 18.9

0.060 0.696 49.5 1.37 6.61 19.6

0.082 0.644 33.1 1.54 5.51 16.4

0.100 0.669 30.4 0.91 5.00 33.6

0.053 2.411 -3.7 1.06 10.62

0.3

0.076 2.983 -6.2 1.33 12.14

1.9

0.070 2.313 -6.5 1.54 9.15 -7.1

0.042 1.701

7.5 0.94 9.50

3.9

0.024 2.472 -2.8 0.45 11.37 -0.9

0.143 5.239 -20.9 1.65 17.76 -7.0

0.241 6.722 -26.7 2.02 18.92 -1.6

0.191 6.699 -29.2 1.39 19.60 -11.1

0.140 3.523

9.5 3.11 18.47 -3.5

-0.001 2.692 13.8 0.09 13.22 -14.1

0.054 1.407 -6.8 1.09 5.75 -7.0

0.022 1.083 -8.3 0.43 3.65 -14.9

0.047 1.636 -12.4 0.97 6.80

3.2

0.091 1.437 -8.6 1.79 5.45 -12.9

0.056 1.415

2.8 1.20 6.51 -14.0

0.079 2.349

4.4 1.65 11.30

2.8

0.088 2.886

8.2 1.76 14.12 12.0

0.085 2.729 -0.7 1.95 11.71 -5.7

0.106 1.918 -0.3 1.95 8.64 -2.5

0.036 1.614 14.8 0.93 9.94

4.9

0.070 2.790 -2.2 1.33 11.65

6.3

0.069 3.103 -6.0 1.06 11.91

3.4

0.060 2.936

4.3 1.30 12.68 19.1

0.060 2.389 -1.0 3.27 10.94 -12.5

-0.001 2.677 -6.0 -0.29 10.69

5.3

0.054 1.660 11.6 1.17 8.18

2.7

0.063 1.866

7.4 1.45 9.04

1.6

0.055 1.710 19.7 1.37 8.96

6.6

0.042 1.600

6.5 0.92 6.75 -7.2

0.042 1.436 13.4 1.03 7.74

3.6

0.043 2.156

0.4 0.86 9.37 -6.5

0.031 1.395 -0.7 0.69 6.42 -15.1

0.064 1.475

6.0 0.71 7.19 -2.2

0.102 1.441

8.6 2.12 4.58 -6.4

-0.025 3.518 -1.8 -0.73 14.18 -8.4

0.039 1.423 -0.4 0.81 6.17

6.6

0.068 1.257

6.2 1.39 5.62

6.9

0.028 1.355

3.8 0.66 5.97 -1.6

0.058 1.375 -6.7 1.10 5.46 14.4

0.002 1.670 -2.8 0.07 7.35

5.1

0.050 3.388 -0.6 0.83 14.88 -12.0

0.014 1.426 10.3 0.32 6.63 15.2

0.062 1.914 12.1 1.53 8.59 -9.9

-0.014 3.051

8.6 -0.43 15.28 -20.7

0.137 5.558 -5.6 1.88 22.98 -11.1

0.072 1.446

5.6 1.51 6.67

2.8

0.026 0.864

5.2 0.56 3.61 -3.8

0.036 1.355 12.4 0.90 7.47 17.6

0.110 1.510

7.2 2.20 6.21 -10.0

0.116 1.868

0.5 2.49 8.25 -5.4

Daily Sampling

Monthly Sampling

^

^ ^(1)

^

^ ^(1)

0.041 0.807 19.3 0.85 4.38

5.8

0.033 0.635 25.9 0.68 3.82

7.1

0.022 0.813 29.4 0.60 4.60

6.6

0.067 0.822 16.0 1.29 4.55 -4.7

0.040 0.822 11.1 0.85 4.49 -7.2

0.075 0.687 38.7 1.08 5.67 22.2

0.066 0.728 38.8 1.03 5.44 20.3

0.057 0.696 49.5 1.16 6.39 21.2

0.079 0.646 33.1 1.38 5.55 15.0

0.097 0.676 30.9 0.77 5.22 31.7

0.024 2.396 -4.1 0.52 10.31

0.2

0.032 2.952 -6.9 0.63 11.58

2.4

0.044 2.295 -6.9 1.13 8.82 -6.6

0.028 1.694

7.4 0.50 9.34

3.7

-0.007 2.470 -3.2 -0.18 11.19 -2.1

0.009 5.155 -21.8 0.19 16.96 -9.7

0.022 6.590 -28.5 0.46 17.14 -5.1

-0.027 6.577 -30.2 -0.25 17.52 -12.4

0.079 3.497

8.4 1.43 18.36 -7.9

-0.038 2.714 15.2 -0.88 14.49 -12.8

0.044 1.405 -6.9 0.94 5.55 -6.8

0.051 1.080 -8.3 0.36 3.60 -14.7

0.034 1.637 -12.5 0.76 6.19

4.4

0.081 1.434 -8.8 1.63 5.31 -13.2

0.046 1.411

2.8 0.99 14.49 -12.4

0.051 2.333

4.1 1.03 10.92

4.2

0.047 2.852

7.8 0.84 13.34 14.3

0.047 2.723 -1.2 1.29 11.18 -5.1

0.088 1.910 -0.2 1.56 8.61 -1.0

0.023 1.601 15.1 0.44 9.94

5.5

0.032 2.768 -2.4 0.66 11.53

5.7

0.022 3.074 -6.2 0.35 12.00

3.6

0.018 2.890

3.7 0.55 12.05 17.6

0.126 2.373 -1.2 2.64 10.94 -12.0

-0.037 2.682 -5.6 -0.87 10.78

5.4

0.040 1.661 11.7 0.83 8.21

3.7

0.046 1.862

7.3 0.89 8.82

2.1

0.041 1.710 19.8 1.05 8.85

7.8

0.042 1.599

6.5 0.69 6.67 -5.9

0.031 1.443 14.1 0.70 8.30

4.9

0.020 2.145

0.3 0.43 9.22 -5.0

0.022 1.391 -0.8 0.49 6.31 -14.0

0.040 1.470

5.8 1.12 6.79 -3.9

0.041 1.431

8.5 1.85 7.03 -6.7

0.031 3.498 -2.1 -1.73 14.02 -7.2

0.029 1.427 -0.4 0.61 6.19

6.9

0.060 1.255

6.2 1.22 5.56

7.5

0.019 1.351

3.8 0.48 5.90 -1.4

0.048 1.370 -6.7 0.95 5.37 14.5

-0.012 1.690 -2.8 -0.20 7.57

5.1

-0.008 3.362 -1.4 -0.13 13.65 -11.7

0.004 1.449 10.0 0.08 6.97 18.0

0.043 1.904 12.1 1.17 8.47 -8.7

-0.060 3.027

7.9 -1.57 15.00 -18.1

-0.014 5.505 -6.6 -0.23 19.83 -12.5

0.061 1.447

5.6 1.29 6.55

2.0

0.022 0.862

5.2 0.50 3.58 -3.2

0.027 1.361 12.3 0.62 7.40 15.7

0.099 1.504

7.1 1.99 6.10 -10.3

0.098 1.873

0.7 2.05 8.08 -5.6

Table 2.3. Statistics for Daily and Monthly Simple and Continuously

Compounded Returns

See the left side of Table 2.3 for the required statistics.

See the right side of Table 2.3. If r denotes net simple return in percent, then

100 log(1 + r=100) is the corresponding continuously compounded return in percent.

2.5.1

2.5.2

PROBLEMS IN CHAPTER 2

See Figure 2.1 for the required plots. Returns are truncated to t in the interval

[;3%; 3%], i.e., returns smaller than ;3% are replaced by a return of ;3%, and returns

larger than 3% are replaced by a return of 3%.

2.5.3

0.8

0.6

Proportion

0.2

0.0

0.0

-3

-2

-1

-3

-2

-1

0.8

0.6

Proportion

0.0

0.2

0.4

0.0

0.2

0.4

0.6

0.8

1.0

1.0

SOLUTION 2.5

Proportion

0.4

0.6

0.4

0.2

Proportion

0.8

1.0

1.0

-3

-2

-1

-3

-2

-1

PROBLEMS IN CHAPTER 2

Use results from Problems 2.5.1 and 2.5.2 (i.e., Table 2.3), counts from Table 2.2, the

assumption that returns are IID, and the asymptotic normality of ^ to obtain estimates

of the 99% condence intervals.

2.5.5 Compute the statistics of interest as in Problem 2.5.1. See also Table 1.1 in the

text for the statistics for the entire sample period. The variances of your estimates can be

estimated via the bootstrap (see Efron and Tibshirani [1993]) under the assumption that

returns are temporally IID. Computing the exact variances for estimators of skewness,

kurtosis, and the studentized range is possible under certain distributional assumptions

for returns, but is quite involved so the bootstrap is the preferred method|it simplies

the estimation greatly: no additional sampling theory is needed). Use the asymptotic

normality of your estimators to perform the tests.

2.5.4

Problems in Chapter 3

Solution 3.1

(S3.1.1)

E[rito ] =

1

X

k=0

E[(rito )2 ] =

=

(S3.1.2)

E[Xit (k)]E[ri;t;k ] =

1

X

k;l=0

1

X

k=0

(1 ; i )ik i = i

k=0

2 ]+

E[Xit (k)]E[ri;t

;k

1X

X

k=0 l>k

1X

X

i2 + 22i

(1 ; i )ik il;k

k=0 l>k

i2 + 22i =(1 ; i ) ;

2

=

1

X

hence

(S3.1.3)

Var[rito ] = E[(rito )2 ] ; 2i = i2 + 2i 2i =(1 ; i )

as in (3.1.10). Next, for n > 0 we have

o ] =

E[rito ri;t

;n

(S3.1.4)

1X

1

X

k=0 l=0

nX

;1 X

1

k=0 l=0

2i (1 ; in ) ;

=

which yields the rst part of (3.1.11). For i 6= j and n 0 we have

o ] =

E[rito rj;t

;n

(S3.1.5)

1X

1

X

k=0 l=0

= i j +

=

1

X

l=0

; j )

2

i j + in (1 ;1;i )(1

i j i j f

and the second part of (3.1.11) follows. Equation (3.1.12) is direct consequence of (3.1.10)

and the rst part of (3.1.11).

9

10

PROBLEMS IN CHAPTER 3

Solution 3.2

Consider the case where the common factor ft is the (observable) market portfolio. Then

the true beta of security i is i as in (3.1.1) and the beta computed from observed returns

io is given by

io = Cov[rito ; ft ]= Var[ft ]

= E[ft

(S3.2.1)

1

X

k=0

= (1 ; i )i :

Thus, the beta will be biased towards 0 if nonsynchronous trading is not properly accounted for.

Solution 3.3

Let Pit and Qit denote the unconditional probabilities that it = 0 and it = 1,

respectively and let Pi and Qi be the corresponding steady-state probabilities. Then

(3.5.1) yields

3.3.1

and similarly for Z. In steady state, Pi = Pit = Pi;t;1 and Qi = Qit = Qi;t;1, hence

0

Pi = 2 ;1(; +i 0 )

(S3.3.2)

i

i

1

;

i

Qi = 2 ; ( + 0 ) :

(S3.3.1)

of it is

i = Qi

i2 = E[it2 ] ; E[it ]2 = Qi (1 ; Qi )

ii (1) = E[it i;t;1 ] ; 2i = Qi(i0 ; Qi) :

3.3.2 To calculate the statistics of observed returns, use (3.1.4). For the mean we have

(S3.3.3)

(S3.3.4)

(S3.3.5)

(S3.3.6)

Ri =

1

X

k=0

E[Xit (k)]E[Ri;t;k ]

!

1

X

0

k

;

1

= i i Qi + Pi i (1 ; i ) = i (i0 Qi + Pi ) ;

k=1

2

Ri

(S3.3.7)

" X

!2#

1

= E

Xit (k)Ri;t;k ; 2Ri

k=0

0

1

1

X

2

0

= i @i Qi +

E[Xit (max(k; l))]A ; 2Ri

Pi k;l=1

= 2i (1 ; ) + i0 Qi ; (i0 Qi + Pi )2 ;

i

i

SOLUTION 3.4

11

21

3

X

ii (1) = E 4 Xit (k)Xi;t+1 Ri;t;k Ri;t;k;1 5 ; 2Ri

k;l=0

1

X

2

0

2

l=0

;

= 2i i0 ; (i0 Qi + Pi )2 :

Thus, serial correlation in it decreases the mean Ri as compared to the case of no

nonsynchronous-trading eects, ceteris paribus, since i0 Qi + Pi < 1 for 0 < i ; i0 < 1.

T

3.3.3 Assume we are given a sequence fit gt=0 of no-trade indicators. For convenience,

we shall condition on the initial no-trade indicator i0 (the extension to the general

case is straightforward). Denote by ni00 ; ni01 ; ni10 , and ni00 the counts of all pairs of

(S3.3.8)

= i

consecutive days with no-trade patterns `00',`01',`10', and `11', respectively. Therefore,

P

nijk = T . Since it follows the Markov process (3.5.1), the log-likelihood function

j;k=0;1

(S3.3.9)

L fit gTt=1 ji0 = ni00 log i + ni01 log(1 ; i ) +

ni10 log(1 ; i0 ) + n11 log i0 :

The maximum likelihood estimators of i ; i0 are

^i = n n+i00n

(S3.3.10)

i00

i01

n

i11

0

^i = n + n :

i10

i11

and the Fisher information matrix is

!#

" ni00 + ni01

0

(1;i )2

i2

(S3.3.11)

i(i ; i0 ) = E

ni10

ni11

0

(1;i0 )2 + i0 2

so that our estimates ^i ; ^i0 are asymptotically independent and normal, with asymptotic

variances estimated eciently by

(S3.3.12)

^2^i = (^i (1 ; ^i )(ni00 + ni01 ));1

;

^2^i0 = ^i0 (1 ; ^i0 )(ni10 + ni11 ) ;1 :

The results of the empirical analysis are given in Tables 3.1 and 3.2. In Table 3.1, the nontrading counts are reported for six securities using ve years of daily data from January 4,

1988 to December 31, 1992. The data was extracted from the CRSP daily master le: out

of 1,120 ordinary common shares continuously listed on the NYSE over this time span,

360 did not trade at least on one of the NYSE trading dates, and 56 did not trade on

at least 100 days out of 1,517 days in total. Our sample is a randomized selection of six

stocks from the latter set. Values for ni0 and ni1 , dened analogously to ni00 etc., are

also provided for convenience. Note that ni01 and ni10 coincide in some cases.

Estimates of i and i0 are given in Table 3.2.

Solution 3.4

Let the Markov process for It be given by the transition probability matrix

(S3.4.1)

C 1 ;p q 1 ;q p ;

with steady-state probabilities of It being ;1 and 1 given by P (1 ; q)=(2 ; p ; q) and

Q (1 ; p)=(2 ; p ; q), respectively.

12

PROBLEMS IN CHAPTER 3

Input Data for Problem 3.1.3. Representative sample from infrequently traded (at

least 100 no-trade days in the sample interval) ordinary common shares continuously

listed on the NYSE from 1988 to 1992. Each stock is identied by its ticker symbol and

CUSIP number. Counts of days a stock did not trade ni0 , did trade ni1 , and patterns

of non-trading for all pairs of consecutive days ni00 , ni01 , ni10 , ni11 are reported.

Ticker CUSIP ni0 ni1 ni00 ni01 ni10 ni11

ZMX

UNF

JII

MBC

ADU

LVI

98991710

90470810

47936810

59478010

02342610

50243910

390

244

220

173

136

117

1127

1273

1297

1344

1381

1400

129

67

70

25

33

35

260

177

150

148

103

82

261

177

150

148

103

82

867

1096

1147

1196

1278

1318

Parameter0 Estimates for Problem 3.1.3. Maximum likelihood estimates of probabilities i , i for representative sample of infrequently traded stock are reported, together

with estimates of their standard deviations .

Ticker ^i

^^i

^i0

^^i0

ZMX

UNF

JII

MBC

ADU

LVI

0.327

0.275

0.318

0.145

0.243

0.299

0.108

0.143

0.145

0.216

0.200

0.202

0.769

0.861

0.884

0.890

0.925

0.941

0.071

0.081

0.087

0.087

0.102

0.114

Then Pt is a four-state Markov process where the quasi-state Pt = 0 is in fact two

distinct states according to whether the pair (It;1 ; It) is (;1; ;1) or (1; 1). In the steady

state, we have the following transition probability matrix:

1

0

0

p2

1;p

2

B@ (1;p)q(1;q) p (1;q)+q (1;p) p(1;p)(1;q) C

(S3.4.2)

p(1;q)+q(1;p)

p(1;q)+q(1;p)

p(1;q)+q(1;p) A :

1;q

q

0

The moments of Pt are then

E[Pt ] = 0;

2

; q)

(S3.4.3)

Var[Pt ] = 2s (12 ;; pp)(1

;q ;

2

; q)

Cov[Pt ; Pt;k ] = ; s (12 ;;pp)(1

;q

C k;1 (1 ; p; q ; 1)0 ; k > 0 :

Corr[Pt ; Pt;k ] = ;(1; ;1)2(2

; p ; q)

Observe that the rst autocorrelation coecient equals ;1=2 as in the IID case, but the

higher-order autocorrelations are nonzero in general.

SOLUTION 3.6

13

Solution 3.5

We will show how discreteness can in
uence and bias several popular stock price statistics. Consider a stock with a virtual price process that follows a continuous geometric

Brownian motion, with a net expected annual return and standard deviation of return

(not continuously compounded) of = 10% and = 20%, respectively.

Assume the observer has available daily-sampled prices rounded to the closest eighth

of a dollar (or to $0.125 if the virtual price is less than $0.125) for a period of ten years.

For purposes of this exercise we neglect the complications of non-trading days and assume

the year consists of 253 equally-spaced trading days.

We shall focus on the estimator ^i of the expected annual returns dened as rescaled

arithmetic average of daily returns, and the estimator ^ of the volatility of annual returns

dened as a rescaled standard deviation of daily returns. The rescaling is as follows: the

average of daily returns is multiplied by thepnumber of trading days 253, and the standard

deviation of daily returns is multiplied by 253. While such estimators might be suitable

for slowly-changing and continuous price processes, they are badly biased estimators of

the theoretical 10% expected return and 20% standard deviation, respectively.

Expressing the parameters of the underlying geometric Brownian motion process as

0 = log p + 1

(S3.5.1)

1 + =( + 1)

(S3.5.2)

0 = log (1 + =( + 1))

and running 4,000 replications of the simulation described above, we report the means of

the statistics for a hypothetical stock with various initial prices in the Table 3.3.

Estimates of return, standard deviation, and autocorrelation are highly biased for

low-priced stocks. Indeed, the hypothetical $0:25 stock exhibits apparent return of almost

50%. For higher stock prices the discreteness biases subside. Nevertheless, we see that

even for high-priced stocks the estimates are still biased due to the way we rescaled daily

estimates to yield annual gures (these estimates would be unbiased if we had assumed

arithmetic instead of geometric Brownian motion).

Problem 3.5 shows that the eects of price discreteness can be substantial for stockreturn statistics and that appropriate care has to be taken to avoid such biases.

Solution 3.6

3.6.1 From the histogram of IBM transaction stock prices on January 4th and 5th, 1988

(Figure 3.1) we observe price clustering around $120 and $123. These clusters correspond

to trades taking place on dierent days.

On the other hand, the histogram of price changes (Figure 3.2) does not exhibit any

apparent clustering, leaving aside the discretization to eighths of dollars (or \ticks"), i.e.,

the smallest price variation possible from one trade to the next. We see that most of

changes fall in the range from ;2 to +2 ticks.

When we compare the two histograms of price changes conditional on prices falling

on an odd or an even eighth (Figure 3.3), we see a dierent pattern: there are fewer

zero-tick price changes that fall on odd eighths than on even eighths, and relatively more

one-tick price changes that fall on odd eighths than on even eighths. Overall, even-eighth

prices are signicantly more frequent then odd-eighth ones. These regularities underscore

the potentially important impact that discreteness can have on statistical inference for

transactions data.

3.6.2 The histogram of times between trades for IBM stock (Figure 3.4) shows that the

majority of trades take place within intervals shorter than one minute. Based on n = 2;746

time intervals, the estimate of the expected time between trades is ^ = 16:86 and the

estimate of the standard deviation of the time between trades is ^ = 19:46. The 95%

14

PROBLEMS IN CHAPTER 3

Histogram for IBMs Stock Price

300

250

Number of trades

200

150

100

50

0

116

117

118

119

120

121

Price in Dollars

122

123

124

Histogram of IBMs Stock Price Changes

1600

1400

Number of Trades

1200

1000

800

600

400

200

0

-6

-4

-2

0

2

Change of Price (Eighths of Dollar)

SOLUTION 3.6

15

Number of Trades

1000

800

600

400

200

0

-6

-4

-2

0

2

Change of Price (Eighths of Dollar)

Number of Trades

1000

800

600

400

200

0

-6

-4

-2

0

2

Change of Price (Eighths of Dollar)

Figure 3.3. Histogram of IBM Price Changes Falling on Odd or Even Eighth

Histogram of Times Between Trades for IBM

500

450

400

Occurrences

350

300

250

200

150

100

50

0

0

20

40

60

80

100

120

140

Time Between Consecutive Trades (sec)

160

180

200

16

PROBLEMS IN CHAPTER 3

a $1/8 grid on naive estimates of annual mean and standard deviation based on daily

returns data is simulated for a hypothetical stock following a continuous time geometric

Brownian motion price process with an annual expected return of 10% and an annual

standard deviation of 20%. For a low-priced stock discreteness biases are substantial.

For a high-priced stock the main source of bias is the misspecication of the process for

purposes of estimation, e.g. taking arithmetic means instead of geometric. The statistics

are based on 4,000 replications of 10 years of daily price data for each row.

Initial price Expected Return Standard Deviation Autocorrelation

0.25

0.4949

(0.0043)

0.9170

(0.0049)

-0.2609

(0.0010)

0.50

0.2893

(0.0012)

0.6454

(0.0021)

-0.2821

(0.0007)

1.00

0.1847

(0.0006)

0.4537

(0.0012)

-0.2886

(0.0005)

2.00

0.1327

(0.0007)

0.3236

(0.0008)

-0.2734

(0.0005)

5.00

0.1038

(0.0008)

0.2210

(0.0003)

-0.1562

(0.0009)

10.00

0.0977

(0.0009)

0.1918

(0.0001)

-0.0562

(0.0006)

20.00

0.0963

(0.0009)

0.1834

(0.0000)

-0.0167

(0.0003)

50.00

0.0969

(0.0009)

0.1808

(0.0000)

-0.0033

(0.0003)

100.00

0.0934

(0.0009)

0.1805

(0.0000)

-0.0010

(0.0003)

condence interval for the expected time can be therefore estimated as

(S3.6.1)

(^ ; 1:96^ =n1=2 ; ^ + 1:96^ =n1=2 ) = (16:23; 17:49) :

Suppose that trade times follow a Poisson process with parameter . That is, assume

that the probability Pk of exactly k trades occurring during any one-minute interval is

given by

k

Pk = e; k! :

(S3.6.2)

The sample average time between trades ^ is a sucient statistic for ; in fact, ^ = 60=^

is a consistent and ecient estimator of . Note that the number 60 is the result of

rescaling time from seconds to minutes. For our sample, ^ = 3:56. We can map the 95%

condence interval (S3.6.1) of derived above into the following 95% condence interval

for

60

60

(S3.6.3)

;

^ + 1:96^ ^ ; 1:96^ = (3:43; 3:69) :

SOLUTION 3.6

17

Price in Dollars

122

120

118

116

9

10

11

12

13

14

Day Time in Hours

15

16

17

16

17

500

400

300

200

100

0

9

10

11

12

13

14

Day Time in Hours

15

Note that this condence interval is not centered on ^, but has the advantage of following so

directly from the condence interval for ^ . As n ! 1, both ^ and ^ are asymptotically

normal consistent estimates of and .

It also follows from the denition of the Poisson distribution that the probability of

no trade during a one-minute interval can be estimated by

(S3.6.4)

Dividing the two trading dates into one-minute intervals and counting the number of

trades, we get a total of 776 minutes (excluding possible opening and closing lags each day).

A trade occurred in 733 of them. Furthermore, 697 minutes in which a trade occurred

were immediately preceded by a minute in which a trade occurred as well. Therefore, the

estimate of the probability of a trade occurring within a particular minute is 0:0554 with a

95% condence interval of (0:0393; 0:0715), and the estimate of the probability of a trade

occurring within a particular minute conditional on a trade occurring in previous minute

is 0:0491 with a 95% condence interval of (0:0335; 0:0648). Estimates of conditional

and unconditional probabilities do not dier statistically signicantly, hence we cannot

reject the hypothesis of independence on these grounds. On the other hand, there is a

statistically signicant discrepancy between these sample probabilities and the estimate

based on the Poisson assumption. Thus, we can reject the independence of trades in that

sense.

3.6.3 Plots of price and volume against time-of-day for both days exhibit certain patterns

(Figures 3.5, and 3.6). Price discreteness is visible from its price path; volume exhibits

18

PROBLEMS IN CHAPTER 3

IBMs Stock Price Path, Jan 5, 1988

Price in Dollars

123.5

123

122.5

122

121.5

9

10

11

12

13

14

Day Time in Hours

15

16

17

16

17

250

200

150

100

50

0

9

10

11

12

13

14

Day Time in Hours

15

large skewness and kurtosis; there is apparently less volume around lunchtime. Time-ofday phenomena are probably untestable from a sample of two days. There is no apparent

relationship between price movements and volume visible by naked eye.

Consider the simple qualitative hypothesis that large-volume trades are accompanied

by price movements of dierent magnitude than small-volume trades. Let us partition

sample of 2,746 trades into nb = 42 block trades (trades that are greater than or equal to

100 round lots) and ns = 2;704 smaller trades, and compute the sample means ^b , ^s and

standard errors of absolute price changes immediately following the trades, expressed in

dollars:

^b = 0:0446 (0:0102)

(S3.6.5)

^s = 0:0675 (0:0019) :

The dierence of these averages is 0:0229 with a standard error of 0:0104, which is significantly dierent from zero at the 5% level. Therefore, trading volume is indeed linked to

subsequent price changes. Note that block trades are followed by smaller price changes

than the majority of small volume trades.

3.6.4 Consider the following simple model for estimating the price impact of selling IBM

stock. Assume that we cannot distinguish whether a trade was \seller-initiated" or \buyerinitiated" from the data, so that we will relate only the absolute magnitude of trading

volume to the absolute magnitude of price change as in the previous part. Moreover, assume that the (absolute) price impact of a trade is proportional to volume, ceteris paribus,

and that errors of measurement are, after division by volume, independent and identically

distributed. Under these strong but simple conditions we can estimate eciently the coecient of proportionality between volume and its price impact as the sample mean ^

of ratios of absolute price changes to volume, according to the Gauss-Markov theorem.

SOLUTION 3.7

19

800

700

Number of Quotes

600

500

400

300

200

100

0

0.1

0.15

0.2

0.25

0.3

0.35

BidAsk Spread in Dollars

0.4

0.45

0.5

In our data ^ = 0:0310 (in dollars per one round lot) with a standard error of 0:0013.

Thus, we can conclude that the seller of one round lot eectively pays three cents per

share less than marginal seller. This amount becomes economically interesting in the case

of block trades where the loss is of the order of $3 per $120-share of stock.

Solution 3.7

January 4th and 5th, 1988, only four sizes of spreads occurred among 1,327 quotes. There

were 748 quotes with a spread of one tick, 502 with two ticks, 65 with three ticks, and 11

with four ticks. A histogram (Figure 3.7) shows that one-tick and two-tick spreads were

by far the most common.

Bid-ask spread dynamics are not IID. Table 3.4 displays the empirical distribution of

bid/ask spreads both unconditionally and conditionally on the previous quote's spread. It

is apparent that the conditional distributions dier signicantly from the unconditional

distribution.

3.7.2 The question of \causality" between quote revisions and transactions is dicult to

answer with the data at hand if we wish to take into account agents' expectations about

future events. Thus, for simplicity we shall consider \causality" strictly in the temporal

sense: does an increase in the spread come before or after an increase in trading volume,

ceteris paribus?

First, for simplicity let us measure intensity of transactions activity at any time

interval by the number of shares traded in that interval, independently of how the volume

is broken up to individual trades and independently of the stock price.

Let us partition the trading day to n = 1; : : : ; N , roughly 15-second intervals delimited by a subset of quotes. Let variables sn , vn+ and vn; indicate changes in quote spread

3.7.1

20

PROBLEMS IN CHAPTER 3

stock during January 4th and 5th, 1988. Relative frequencies of bid/ask spreads conditional on preceding quote's spread are expressed in percent. Spreads are denominated

in ticks.

Previous

Current Spread

Spread

1

2

3

4

1

2

3

4

Any

71.4

40.0

18.5

9.1

27.1 1.5

51.6 7.2

52.3 23.1

54.5 27.3

56.4 37.9

4.9

0.0

1.2

6.2

9.1

0.8

and transaction volume related to nth interval. More specically, let sn be UP, if quote

spread increases between quotes delimiting nth interval, UNCH if spread does not change,

and DOWN if spread decreases. Also, vn+ is UP if trading volume at interval n is smaller

than that in n + 1, etc. Analogously, vn; be UP if trading volume in n ; 1 is smaller than

that in n.

Estimates of joint probabilities of s, v+ and v; allow some statistical inference about

the relation between spreads and transactions. In particular, if quote revisions aect only

subsequent transactions but do not in
uence previous one, the variables s and v; should

be statistically independent. On the other hand, if quotes re
ect previous transaction

activity, s and v+ should be statistically independent. The empirical distribution of the

27 triples [v; ; s; v+ ], under assumption that their realizations at triples of consecutive

15-second intervals are IID, allow us to test the proposed hypotheses.

However, it is well possible that transaction activity and quote revisions in
uence

temporally each other. That being the case, testing existence of unilateral causality may be

next to meaningless. Therefore, let us test \causality" in each direction separately, against

alternative hypothesis of no relation between quote revisions and transaction activity. In

another words, let us test whether variables s and v+ are dependent, to see whether current

quote revisions in
uence future transactions. Similarly, let us test whether variables s and

v; are dependent, to see whether current quote revision is in
uenced by past transactions.

Using a standard asymptotic test of independence for a contingency table as described

in Rao (1973, pp. 404{412), we have, under the null hypothesis of independence, that:

r X

s

X

(nij ; ni: n:j =n:: )2

2(r;1)(s;1) =

(S3.7.1)

ni: n:j =n::

i=1 j=1

has 2 distribution with (r ; 1)(s ; 1) degrees of freedom. In our case r = s = 3 we have

24 . The contingency tables (Tables 3.5) provide a summary of the data.

It turns out that 24 statistics for the rst table is 13.6, and for the second 17.9 so that

we reject the hypothesis of no dependence between s and v; on 0.9% signicance level

and that of s and v+ on 1.2% signicance level. Thus we have shown that quote revisions

\in
uence" future transactions, and past transactions \in
uence" quote revisions.

The next step of the analysis may be to postulate a particular model that involves

both eects between transactions and quotes, and perform another round of the statistical

analysis.

3.7.3 This part is very similar to 3.7.2, hence we omit the solution.

3.7.4 Let us assume that the investor starts with the bond position and that considers

the quotes as the relevant price information sense: the investors account only for the

SOLUTION 3.7

21

Contingency Tables for causal relationship between transactions activity and quote

revisions. The trading day Jan 4, 1988 is divided to quote-to-quote intervals of roughly

15 seconds apart and changes of bid/ask spread together with changes in the trade

volume in these intervals are counted. Tables show relationship between past/future

changes in transactions activity against the spread change in the intervals.

Past

Current Spread

Future

Current Spread

Trading UP UNCH DOWN

Trading UP UNCH DOWN

UP

UNCH

DOWN

23

3

21

59

12

50

10

2

32

UP

UNCH

DOWN

20

2

25

51

2

68

33

2

9

rise or decline of the mid-price given as average of the bid/ask spread. Quotes that do

not change the mid-price are eectively ignored. Further assume that at the end of the

two-day trading period the stock position is liquidated into bonds.

A simulation of such a trading strategy shows: (1) if the investor is allowed to buy

and sell at the average, he is left with $101,899 at the end; (2) if the bid/ask prices are

used, he is left with $97,769. We see that the bid/ask spread does matter.

It would be dicult to perform any sensible statistical analysis based on the one

simulation performed. In particular, it is incorrect to assert that the strategy in (1) that

led to nearly a 1% return over one day would dominate a buy-and-hold strategy for a

dierent data set or over a dierent time span. Nevertheless, the gap between the prots

of (1) and (2) are real: frequent trading and large spreads do create signicant losses

compared to the \frictionless" case.

22

PROBLEMS IN CHAPTER 3

Problems in Chapter 4

Solution 4.1

for ^i from (4.5.7) we have ^i = Ri ; Xi ^i ! Ri ; Xi i = i in probability, as L1 ! 1.

Because abnormal returns i are independent (across time), the sample abnormal returns

^i are asymptotically independent as L1 ! 1.

Solution 4.2

We assume that the cumulative abnormal return test statistics are calculated using the

known standard deviation of the abnormal returns, that the abnormal returns are independent through time and across observations and normally distributed, and that the

abnormal returns are measured without parameter sampling error (L1 is large). Denote

L2 = 3 as the length of the event window and N as the number of event observations.

Designate group 1 as the observations with low standard deviation and group 2 as the

observations with high standard deviation. For the group means and standard deviations

we have 1 = 0:003, 2 = 0:003, 1 = 0:03, and 2 = 0:06 where the subscript indicates

the group. N1 = 25 and N2 = 25 are the number of observations in groups 1 and 2

respectively.

To calculate the power against the given alternative, we need to derive the distributions of the test statistics under that alternative. First, we aggregate the abnormal returns

over the event window for each observation which gives

(S4.2.1)

CARi =

L2

X

l=1

il :

Given the assumptions, E[CARi ] = L2 g(i) and Var[CARi ] = L2 g(i)

the group of observation i.

Then, we aggregate across observations to form the test statistics (modied to re ect

the above assumptions). The aggregation of abnormal returns corresponding to J1 in

(4.4.22) is

(S4.2.2)

J1 =

N

1

N2

i=1

(S4.2.3)

CARi ;

N CAR

X

p

q i 5:

J2 = N 4 1

i=1

23

2

L2 g(i)

24

PROBLEMS IN CHAPTER 4

Under the specied alternative hypothesis, the distributions of the test statistics J1

and J2 are

p

2 ) ; 1);

J1 N (1 ; 1) = N ( (NL2 (N2 1+1N+N2 )21=2

(S4.2.4)

1 1

2 2

N

(S4.2.5)

J2

Substituting in the alternative parameter values, for the means of J1 and J2 we have

1 = 0:775 and 2 = 0:919, respectively.

Consider a two-sided test of size based on J1 and J2 , respectively, of the null

hypothesis H0 : [1 2 ] = [0 0] against the alternative hypothesis HA : [1 2 ] =

[0:003 0:003]. Using equation (4.6.1), the powers of the tests, P1 and P2 are

P1 = Pr[J1 < ;1 (=2)] + Pr[J1 > ;1 (1 ; =2)]

(S4.2.6)

= [(;1 + ;1 (=2))] + [1 ; (1 + ;1 (1 ; =2))];

P2 = Pr[J2 < ;1 (=2)] + Pr[J2 > ;1 (1 ; =2)]

= [(;2 + ;1 (=2))] + [1 ; (2 + ;1 (1 ; =2))]:

Evaluation of these expressions for = 0:05 gives P1 = 12:1% and P2 = 15:1%.

Solution 4.3

The solution is the same as for Problem 4.2 except that 2 = 0:006 instead of 0:003. Using

this value for 2 , we have 1 = 1:162 and 2 = 1:225 giving P1 = 21:3% and P2 = 23:3%.

Problems in Chapter 5

Solution 5.1

(S5.1.1)

Ra = 0 + 1 Rop + 2 Rp + p

using well-known regression results, we have

(S5.1.2)

1 = Cov[Ra ; Rop ]= Var[Rop ] = aop;

(S5.1.3)

2 = Cov[Ra ; Rp ]= Var[Rp ] = ap ;

(S5.1.4)

0 = a ; (aop op + ap p );

since Cov[Rp ; Rop ] = 0. The result 2 = ap is immediate, thus we need to show that

1 = 1 ; ap and 0 = 0 to complete the solution.

Let r be the minimum variance portfolio with expected return equal to that of portfolio a, a = r . From the form of the solution for the minimum variance portfolio weights

in (5.2.6), Rr can be expressed as

(S5.1.5)

Rr = (1 ; )Rop + Rp

where = (r ; op )=(p ; op ). Using Cov[Rp ; Rop] = 0 and r = (1 ; )op + p we

have

rop = Cov[Rr ; Rop ]= Var[Rop ]

= Cov[(1 ; )Rop + Rp ; Rop ]]= Var[Rop ]

(S5.1.6)

= (1 ; )

(S5.1.7)

rp = Cov[Rr ; Rp ]= Var[Rp ]

= Cov[(1 ; )Rop + Rp ; Rp ]]= Var[Rp ]

=

(S5.1.8)

r = ropop + rp p :

Portfolio a can be expressed as portfolio r plus an arbitrage (zero-investment) portfolio

a composed of portfolio a minus portfolio r (long a and short r). The return of a is

(S5.1.9)

Ra = Ra ; Rr :

Since a = r , the expected return of a is zero. Because a is an arbitrage portfolio

with an expected return of zero, for any minimum variance portfolio q, the solution to the

optimization problem

min

Var[Rq + cRa ]

(S5.1.10)

c

25

26

PROBLEMS IN CHAPTER 5

is c = 0. Any other solution would contradict q being minimum variance. Noting that

Var[Rq + cRa ] = Var[Rq ] + 2cCov[Rq ; Ra ] + c2 Var[Ra ] we have

@

(S5.1.11)

@c Var[Rq + cRa ] = 2Cov[Rq ; Ra ] + 2c Var[Ra ]:

Setting this derivative equal to zero and substituting in the solution c = 0 gives

(S5.1.12)

Cov[Rq ; Ra ] = 0:

Thus the return of a is uncorrelated with the return of all minimum variance portfolios.

Using this result we have

Cov[Ra ; Rp ] = Cov[Rr + Ra ; Rp ]

(S5.1.13)

= Cov[Rr ; Rp ]

Cov[Ra ; Rop] = Cov[Rr + Ra ; Rop]:

(S5.1.14)

= Cov[Rr ; Rop ]

From (S5.1.13) and (S5.1.14) it follows that

(S5.1.15)

aop = rop

(S5.1.16)

ap = rp :

Combining (S5.1.2) with (S5.1.6), (S5.1.7), (S5.1.15) and (S5.1.16) we have 1 = aop =

1 ; ap . Since r = a , combining (S5.1.4) with (S5.1.8), (S5.1.15), and (S5.1.16) gives

0 = 0 which completes the solution.

Solution 5.2

Begin with the excess return market model from (5.3.1) for N assets. Taking unconditional

expectations of both sides and rearranging gives

(S5.2.1)

= ; m :

Given that the market portfolio is the tangency portfolio, from (5.2.28) we have the (N 1)

weight vector of the market portfolio

(S5.2.2)

!m = 0

1;1

;1 :

Using !m we can calculate the (N 1) vector of covariances of the N asset returns with

the market portfolio return, the expected excess return of the market, and the variance of

the market return,

Cov[Z; Zm ] =

! m = 0

1;1

(S5.2.3)

(S5.2.4)

(S5.2.5)

0 ;1

m = !0m = 0

;1

0 ;1

Var[Zm ] = !0m

!m = (0

;1 )2 :

Z; Zm ]

0

;1

(S5.2.6)

m = Cov[

Var[Zm ] = 0

;1 :

and combining (S5.2.6) and (S5.2.4) we have

(S5.2.7)

m m = :

From (S5.2.1) and (S5.2.7) the result = 0 is immediate.

SOLUTION 5.4

27

Solution 5.3

The solution draws on the statistical analysis of Section 5.3. The calculations for three

selected stocks are left to the reader.

Solution 5.4

Let Zt be a (N +1 1) vector of excess asset returns with mean and covariance matrix

is full rank. (If the

market portfolio is a combination of the N included assets, this assumption can be met

by eliminating one asset.)

From (5.2.28) the tangency portfolio q of these N + 1 assets has weight vector

!q = 0

;1 1

;1 :

(S5.4.1)

Using straight forward algebra we have

2q = (!0q )2 = 0

;1

(S5.4.2)

q2 !0q

!q

The covariance matrix

can be partitioned in the rst N assets and the market portfolio,

2

2 3

m

5

(S5.4.3)

4

0 m2 m2

(S5.4.4)

2 0 2 + 2 3

m

m

5

(S5.4.5)

4

0 m2

m2

where

= 0 m2 + is substituted.

Using the formula for a partitioned inverse (see Morrison (1990) page 69) we have

;1

;1

(S5.4.6)

;1

0

;

1

0

;

1

1

m2 +

(S5.4.7)

Using 0 = [0 m ] and (S5.4.6) we have

2

0

;1 = m2 + ( ; m )0 ;1 ( ; m ):

(S5.4.8)

m

Substituting = ; m gives

2

(S5.4.9)

0

;1 = m2 + 0 ;1 :

m

From (S5.4.2) and (S5.4.9) we have

2

2

0 ;1 = q2 ; m2

(S5.4.10)

q

m

which is the result in (5.5.3).

28

PROBLEMS IN CHAPTER 5

Problems in Chapter 6

Solution 6.1

Let the number of portfolios in the set be K and let RKt be the (K 1) vector of

time period t returns for the portfolios. Since the entire minimum variance boundary can

be generated from the K portfolios, for any value of the constant y, there exists a combination of the portfolios with expected return y which is minimum variance with respect

to the K portfolios plus the N assets. Choose y to be any value but the global minimum

variance portfolio expected return (see equation (5.2.11)) and denote this portfolio op.

Corresponding to op is a minimum variance portfolio p whose return is uncorrelated with

the return of op (see Section 5.2). Since p and op are minimum variance portfolios their

returns are linear combinations of the elements of RKt ,

(S6.1.1)

Rpt = R0Kt !Kp

(S6.1.2)

Ropt = R0Kt !Kop ;

where !Kp and !Kop are (K 1) vectors of portfolio weights. Because p and op are uncorrelated minimum variance portfolios, we have

(S6.1.3)

= op + p (p ; op )

where

p = Cov[R2t ; Rpt ]

p

1

= 2 Cov[Rt ; R0Kt !Kp ]

p

1

= 2 Cov[Rt ; R0Kt ]!Kp :

(S6.1.4)

p

(S6.1.5)

= op + Cov[Rt ; R0Kt ]! Kp (p ;2op ) :

p

Analogous to (S6.1.5) for the K portfolios we have

K = op + Cov[RKt ; R0Kt ]!Kp (p ;2op ) :

(S6.1.6)

p

Rearranging (S6.1.6) gives

(S6.1.7)

!Kp (p ;2op ) = Cov[RKt ; R0Kt ];1 (K ; op ):

p

Substituting (S6.1.7) into (S6.1.5) gives

(S6.1.8)

= op + Cov[Rt ; R0Kt ]Cov[RKt ; R0Kt ];1 (K ; op ):

Now consider the multivariate regression of N assets on K factor portfolios,

(S6.1.9)

Rt = a + BRKt + t

29

30

PROBLEMS IN CHAPTER 6

coecients, and t is the time period t residual vector. From regression theory we have

(S6.1.10)

B

= Cov[Rt ; R0Kt ]Cov[RKt ; R0Kt ];1

(S6.1.11)

a

= ; BK :

From (S6.1.8) and (S6.1.10), we have

= op + B(K ; op)

(S6.1.12)

= ( ; B)op + BK

Since (S6.1.12) holds for dierent values of op it must be the case that ( ; B) = 0,

that is the factor regression coecients for each asset, including asset a, sum to one. If

( ; B) = 0, then (S6.1.12) reduces to = BK and thus from (S6.1.11) we have a = 0,

that is the regression intercept will be zero for all assets including asset a.

Solution 6.2

Let and

be the mean excess return vector and the covariance matrix respectively

for the N assets and portfolio p,

2 3

5

0 4

(S6.2.1)

(S6.2.2)

(S6.2.3)

(S6.2.4)

p

2

0 2 3

p

p2

p2

2 0 2 + 0 2 3

p

p

5

4

2

2

p

where

= 0 p2 + and = 0 h2 + I2 .

p

;1 which is the

squared Sharpe ratio of the tangency portfolio. As demonstrated in problem 5.4, given

= a + p and

= 0 p2 + this ratio can be expressed as

2

s2I = 0

;1 = p2 + a0 ;1 a

(S6.2.5)

p

where s2I is the maximum squared Sharpe ratio for economy I , I = A; B: Analytically

inverting = 0 h2 + I2 and simplifying, s2I can be expressed as

2 0 2

(S6.2.6)

s2I = s2p + 12 a0 a + (2 h+(a2 ) 0 ) :

h

2

where sp is the squared Sharpe ratio of portfolio p.

Solution 6.3

Using (S6.2.6) and the cross-sectional distributional properties of the elements of a and ,

an approximation for the maximum squared Sharpe measure for each economy can be derived. For both economies, N1 a0 a converges to a2 , and N1 0 converges to a2 . For economy

A, N12 (a0 )2 converges to a4 , and for economy B , N1 (a0 )2 converges to a4 . Substituting

these limits into (S6.2.6) gives approximations of the maximum squared Sharpe measures

squared for each economy. Substitution into (S6.2.6) gives

a2

(S6.3.1)

s2A = s2p + 2 +N

Nh2 a2

SOLUTION 6.3

31

2

2 2

:

a

s2B = s2p + N a2 1 ; 2 +hN

2

2

h a

Thus we have the squared Sharpe ratios for economies A and B , respectively. The squared

Sharpe ratios for large N follow from (S6.3.1) and (S6.3.2). For economy A we have

(S6.3.3)

s2A = s2p + 12 ;

h

and for economy B we have

2

(S6.3.4)

s2B = s2p + N a2 :

The maximum squared Sharpe measure is bounded as N increases for economy A and

unbounded for economy B . Examples of economies A and B are discussed in section

6.6.3.

(S6.3.2)

32

PROBLEMS IN CHAPTER 6

Problems in Chapter 7

Solution 7.1

Each period, the corporation repurchases shares worth X while the total stock is worth

V X=(1 ; (1 + R);1 ) = (1 + R)X=R. Therefore, the number of shares outstanding

follows the \law of motion"

R

Nt+1 = 1 ; 1 + R Nt :

(S7.1.1)

Price per share is Pt V=Nt , and dividend per share is Dt (1 ; )X=Nt+1 .

(Note that dividends are paid after repurchases, on the remaining shares only). Hence the

growth rate of dividends per share, G, satises

7.1.1

1 + G = NNt = 1 +1R+(1R; ) :

t+1

(S7.1.2)

Dividends per share grow, even though total dividends do not, because the number of

shares is shrinking over time.

7.1.2 The dividend-price ratio is

t+1

(S7.1.3)

DP = R1 +; RG ;

(S7.1.4)

which is consistent with equation (7.1.9) after accounting for the fact that prices here are

cum-dividend, whereas the discussion in the text applies to ex-dividend prices.

7.1.3 This follows immediately from the results of the previous subsection, because if price

is the present value of discounted future dividends, including the dividend paid today, then

(S7.1.5)

Pt =

1

1 D

X

t+i = X D (1 + G)i = D 1 + R ;

t (1 + R)i

t R;G

(1 + R)i

t=0

t=0

Intuitively, shares must have the same value to shareholders who sell shares to the

repurchasing rm and to shareholders who do not. A shareholder who sells a fraction

of his shares to the rm each period receives a constant fraction of total dividends and

repurchase payments, that is, a constant fraction of the rm's cash ow. A shareholder

who sells no shares to the rm receives a growing fraction of total dividends, because the

total number of shares is shrinking over time. The value of the shares is the same in either

case.

33

34

PROBLEMS IN CHAPTER 7

Solution 7.2

1 D

1

X

] = X E [edt +Pnk=1 (+t+k ;r) ]

Ft = Et [

(S7.2.1)

t

;t

=t+1 R

n=1

7.2.1

= Dt

1

X

n=1

2

en(+ =2;r) :

The condition + 2 =2 < r is necessary for the sum to converge. It follows that the ratio

of fundamental value to dividend is

Ft = e+2=2;r :

(S7.2.2)

Dt 1 ; e+2 =2;r

7.2.2 Since

+ Dt+1 ]

(S7.2.3)

Ft + cDt = e;r E[Ft+1 + cDt+1

and

(S7.2.4)

Ft = e;r Et [Ft+1 + Dt+1 ];

we have

]:

(S7.2.5)

Dt = e;r Et [Dt+1

Since

] = eEt [dt+1 ]+Vart [dt+1 ]=2 = e+dt +2 2 =2 ;

(S7.2.6)

Et [Dt+1

we get a quadratic equation for the parameter ,

(S7.2.7)

2 2 =2 + ; r = 0:

For such a parameter , the price process Pt = Ft + cDt indeed gives the same expected

rate of return as the process Pt = Ft .

7.2.3 The Froot-Obstfeld bubble requires a very specic dividend process. However, the

bubble is strongly correlated with the dividend, capturing the eect of dividend \overreaction". The bubble never bursts for a strictly positive dividend stream.

Solution 7.3

Cov[

rt; rt+k]

t

t+k

(S7.3.1)

= E xt;1 + d;t ; 1 ;

x

+

;

t+k;1

d;t+k 1 ;

t x

= E xt;1 xt+k;1 ; 1 ;

t+k;1 :

P

n

Because xt = 1

n=0 t;n , we have

k 2

(S7.3.2)

E[xt;1 xt+k;1 ] = 1 ; 2

and

(S7.3.3)

E[t xt+k;1 ] = k;1 2 :

Thus the return autocovariance is

Cov[rt ; rt+k ] = k;1 1 ;2 ; 1 ; 2 :

(S7.3.4)

7.3.1

This is negative when < . The autocorrelation of stock returns is determined by the

balance of two opposing eects. Expected stock returns are positively autocorrelated,

SOLUTION 7.4

35

and this creates positive autocorrelation in realized stock returns. However innovations

in expected future stock returns are negatively correlated with current unexpected stock

returns, and this creates negative autocorrelation in realized stock returns. The latter

eect dominates when < .

7.3.2 Assume now that

(S7.3.5)

Cov[d;t ; t] = ; > 0:

We have

!

;

k

;

1

2

Cov[rt ; rt+k ] =

(S7.3.6)

2 + 1 ; 2 ; 1 ; :

If ; is large enough, the rst term can dominate the others, giving positive return

autocovariances.

Solution 7.4

(S7.4.1)

rt+1 k + pt+1 + (1 ; )dt+1 ; pt = k + + t+1 :

We see that rt+1 is just a constant plus a white noise component | the log stock return

rt+1 is therefore unforecastable.

7.4.2 Let us rewrite the formula for vt and substitute in the dividend rule. We get

vt ; dt = 1 ; (vt;1 ; dt;1 ) + ; c ; t + t ;

(S7.4.2)

so the log dividend-price ratio dt ; vt follows an AR(1) process with persistence coecient

(1 ; )=.

7.4.3 The log dividend-price ratio is

(S7.4.3)

dt ; pt = dt ; (vt ; (dt ; vt )) = (1 + )(dt ; vt ):

7.4.1

multiple of dt ; vt , it is also an AR(1) process.

The approximate log stock return can be rewritten, using the formulas for dt and pt,

as

rt+1 k + (pt+1 ; dt+1 ) + (dt+1 ; pt)

= k + (1 + )(vt+1 ; dt+1) + c + (1 ; + )(dt ; vt ) + t+1

(S7.4.4)

= constant + 1

+ (dt ; pt ) ; t+1 + (1 + )t+1 :

Setting xt ( =(1 + ))(dt ; pt), we get a model of the form (7.1.27) and (7.1.28), with

xt being the optimal forecasting variable for rt+1 up to a constant.

7.4.4 From the above we have

(S7.4.5)

rt+1 ; Et [rt+1 ] = ; t+1 + (1 + )t+1 ;

and

(S7.4.6)

xt+1 ; Et [xt+1 ] = t+1

;

;

t+1

so that the covariance of the innovations is always negative.

36

PROBLEMS IN CHAPTER 7

Solution 7.5

Let us denote the expectation conditional on the full information set at time t as Et []

and the expectation conditional on information Jt as EJt []. Thus, we have E[Et []] =

E[], EJt [Et []] = EJt [], and so forth, by the law of iterated expectations. In particular,

E[pt ] = E[Et [pt ]] = E[pt ]. Note that the following \prices" are expectations listed in order

of decreasing conditioning information: pt , pt = Et [pt ], p^t = EJt [pt ], and E[pt ].

7.5.1 Calculate

Var[pt ] = E[((Et [pt ] ; EJt [pt ]) + (EJt [pt ] ; E[pt ]))2 ]

(S7.5.1)

= E[Et [pt ] ; EJt [pt ])2 ] + E[(EJt [pt ] ; E[pt ])2 ]

E[(EJt [pt ] ; E[EJt [pt ]])2 ]

= Var[^pt ]:

where the cross term at the second step was eliminated using the fact that

(S7.5.2)

E[EJt [(Et [pt ] ; EJt [pt ])(EJt [pt ] ; E[pt ])]] = 0

as EJt [pt ] ; EJt [pt ] conditional on Jt is a constant and EJt [Et [pt ] ; EJt [pt ]] = 0. Calculations in other parts of the problem are similar.

Intuitively, a price forecast based on less information is less volatile.

7.5.2 Calculate

Var[pt ; p^t] = E[((pt ; pt ) + (pt ; p^t))2 ]

(S7.5.3)

= E[(pt ; pt )2 ] + E[(pt ; p^t )2 ]

= Var[pt ; pt]:

It follows that

(S7.5.4)

Var[pt ; p^t ] Var[pt ; p^t ]

and

(S7.5.5)

Var[pt ; p^t ] Var[pt ; p^t ]

as was to be shown.

Intuitively, a forecast based on inferior information has a larger error variance. Also,

the error variance for a forecast of the actual realization of an uncertain price is larger

than the error variance for a forecast of a superior-information forecast.

Stock prices, referred to in Problem 7.5.1, are usually considered nonstationary so

that their conditional variances do not converge to a nite unconditional variance. On

the other hand, forecast errors, referred to in Problem 7.5.2 may plausibly be assumed

stationary. Therefore, the framework of Problem 7.5.2 seems more suitable for econometric

analysis.

7.5.3 Note that p

^t+1 is dened to be EJt [pt+1 ]. Using the approximation

(S7.5.6)

E[rt+1 ] = E[EJt [rt+1 ]] = E[^rt+1 ]

that follows from (7.1.19) we get

Var[rt+1 ] = E[((rt+1 ; r^t+1) + (^rt+1 ; E[^rt+1 ]))2 ]

(S7.5.7)

= E[(rr+1 ; r^t+1 )2 ] + E[(^rt+1 ; E[^rt+1 ])2 ]

Var[^rt+1 ]:

Intuitively, the variance of a return forecast is less volatile than the return itself. Just

as in Problem 7.5.2, this result is more useful than that in Problem 7.5.1 because the

stochastic processes for returns do not seem to have the unit roots characteristic of price

processes.

Problems in Chapter 8

Solution 8.1

8.1.1 Recall that M =

M

2

2

2

0

0

;

1

(S8.1.1) E[M (M ) ] = M +

= M + ( ; M E[ + Rt ])

( ; M E[ + Rt ]):

t

First, we will show that in the market augmented by a risk-free asset with return

1 + RFt = 1=M , there exists a benchmark portfolio with return

(S8.1.2)

1 + Rbt = Mt(M ) 2 :

E[Mt (M ) ]

Consider a portfolio with dollar weights M on the risky assets and M 2 ; E[ + Rt ]0 M

on the risk free asset. Such a portfolio has payo Mt (M ) and value

0 M + M 2 ; E[ + Rt ]0 M

= M 2 + ( ; M E[ + Rt ])0

;1 ( ; M E[ + Rt ])

(S8.1.3)

= E[Mt (M )2 ]:

Thus the portfolio return is exactly 1 + Rbt , and the proof of existence is complete.

Next, consider any portfolio Rpt such that E[Rpt ] = E[Rbt ]. The properties of Mt (M )

and Rbt imply that

Var[Rpt ] = E[((Rpt ; Rbt ) + (Rbt ; E[Rbt ]))2 ]

(S8.1.4)

= E[(Rpt ; Rbt )2 ] + E[(Rbt ; E[Rbt ])2 ]

= Var[Rpt ; Rbt ] + Var[Rbt ]:

The only nontrivial step was to eliminate the cross term

E[(Rpt ; Rbt )(Rbt ; E[Rbt ])]

(S8.1.5)

= E[Rpt ; Rbt ] Mt(M ) 2 ; E[Rpt ; Rbt ]E[1 + Rbt ]

E[Mt (M ) ]

= 0:

Thus Var[Rpt ] Var[Rbt ] whenever E[Rpt ] = E[Rbt ], so the benchmark portfolio is on the

mean-variance frontier.

8.1.2 First note that E[(Mt (M ) ; Mt (M ))(1 + Rpt )] = 0 for any portfolio return Rpt . It

follows that

(S8.1.6)

Cov[Mt (M ); Rpt ] = Cov[Mt (M ); Rpt ] Var[Mt (M )]1=2 Var[Rpt ]1=2 ;

where equality is attained iff Mt (M ) is perfectly correlated with Rpt . In particular, we

have Cov[Mt (M ); Rbt ] = Var[Mt (M )]1=2 Var[Rbt ]1=2 and therefore

1=2

(S8.1.7)

Corr[Mt (M ); Rpt ] Var[Mt (M )]1=2 = Corr[Mt (M ); Rbt ];

Var[Mt (M )]

37

38

PROBLEMS IN CHAPTER 8

so that Rbt is a maximum correlation portfolio among all Rpt 's with respect to any stochastic discount factor Mt (M ).

8.1.3 From E[Mt (M )( + Rit )] = 1 we get

(S8.1.8)

Cov[Mt (M ); Rit ] = 1 ; E[Mt (M )]E[1 + Rit ]

and similarly for Rbt . Therefore

Cov[Mt (M ); Rit ] = 1 ; E[Mt (M )]E[1 + Rit ] :

(S8.1.9)

Cov[Mt (M ); Rbt ] 1 ; E[Mt (M )]E[1 + Rbt ]

Note that E[Mt (M )] = M and Mt (M ) = c(1 + Rbt ), where c = E[Mt (M )2 ];1 > 0 is a

constant. Thus, the above expression simplies to

Cov[Rbt ; Rit ] 1=M ; E[1 + Rit ]

(S8.1.10)

Cov[Rbt ; Rbt ] = 1=M ; E[1 + Rbt ]

which yields (8.1.17).

8.1.4 Indeed,

M

E[1 + Rbt ] = E Mt(M ) 2 =

(S8.1.11)

;

E[Mt (M ) ]

E[Mt (M )2 ]

and

!1=2

2

(M )(1 + Rbt )]

M

E[

M

t

; ;E[M (M )2 ]

(Rbt) =

E[Mt (M )2 ]

t

(S8.1.12)

so that

(S8.1.13)

Similarly,

(S8.1.14)

E[Mt (M )2 ] ; M 2

E[Mt (M )2 ]

1=2

(Rbt ) = E[Mt (M )2 ] ; 1

E[1 + Rbt ]

M2

1=M ; E[1 + Rbt ] =

(Rbt )

=

Note that

(S8.1.15)

and

1=2

E[Mt (M)2 ] ; M

M2

1=2

E[Mt (M ) ] ; M 2

E[M (M )2] 1=2

t

;1 :

2

8.1.5

E[Mt (M )] = M = E[Mt (M )]

;

(S8.1.16)

= E[Mt (M )2 ] + E[ Mt (M ) ; Mt (M ) 2 ]

E[Mt (M )2 ]:

;

;

Therefore Mt (M ) Mt (M ) and nally

;M (M ) ;M (M )

(

R

)

t

t

bt

(S8.1.17)

E[1 + Rbt ] = E[Mt (M )] E[Mt (M )] :

SOLUTION 8.2

39

Solution 8.2

1 C 1; X

1

X

(S8.2.1)

u (fC g) = j 1 t+j

=

j v(C );

;

j=0

j=0

and consider the maximization problem

(S8.2.2)

max E [u (fC g)]

fC g;fwg t

8.2.1

Consider a single intertemporal sub-problem involving incremental investment of an

amount x in a specic asset i from period t to t + 1, at the cost of time-t consumption,

the proceeds of the investment to be consumed at t + 1:

(S8.2.3)

max

E [v(Ct ; x) + v(Ct+1 + x(1 + Ri;t+1 ))]:

x t

At the optimum of the previous problem, x = 0 has to be optimal here, so that

@ Et [v(Ct ; x) + v(Ct+1 + x(1 + Ri;t+1 ))] = 0;

(S8.2.4)

x=0

@x

which implies

; ] = 0;

(S8.2.5)

Et [Ct; + (1 + Ri;t+1 )Ct+1

from which (8.2.3) follows. Note that (8.2.3) is not only necessary but also sucient for

the optimum once it holds for all i's and t's.

Assuming that asset returns and consumption are jointly log-normal, the quantity

(S8.2.6)

(1 + Ri;t+1 )(Ct+1=Ct );

is also log-normal and therefore by taking logs of (8.2.3)

(S8.2.7)

;

;

Et [log (1 + Ri;t+1 )(Ct+1=Ct ); ] + 21 Vart [log (1 + Ri;t+1 )(Ct+1=Ct ); ] = 0;

so that

(S8.2.8)

Et [ri;t+1 ] + log ; Et [ct+1 ] +

1 ;Var [r ] + 2 Var [c ] ; 2 Cov [r c ] = 0

t i;t+1

t t+1

t i;t+1 t+1

2

which gives (8.2.5).

Assuming that conditional variances and covariances

Vii = Vart [ri;t+1 ];

(S8.2.9)

Vcc = Vart [ct+1 ];

Vic = Covt [ri;t+1 ; ct+1 ]

are all constants, we can write (8.2.5) as

;

Et [ri;t+1 ] = Et [ct+1 ] + ; log ; 21 Vii + 2 Vcc ; 2 Vic ;

(S8.2.10)

which is a linear function of Et [ct+1 ] with slope coecient |the coecient of risk

aversion for the power utility function. This solves part (i). Subtracting (8.2.6), the

riskfree asset equation, we get as in (8.2.7)

(S8.2.11)

Et [ri;t+1 ; rf;t+1 ] + 12 Vii = Vic;

so that the \premium" of the asset is proportional to the conditional covariance of the log

asset return with consumption growth, with coecient of proportionality . This solves

part (ii).

40

PROBLEMS IN CHAPTER 8

Part (i). Let aggregate equity e pay a log dividend equal to log aggregate consumption, so that de;t = ct . From the previous part we know that Et [re;t+1+j ] =

Et [ct+1+j ], up to a constant. Then (7.1.25) implies that

(S8.2.12)

8.2.2

1

X

j=1

By simple algebraic manipulation of the process for ct+1 , we obtain the following

expression for ct+1+j :

(S8.2.13)

ct+1+j =

j !

X

i

j

X

i=0

i=0

+ j+1 ct +

(S8.2.14)

Et ct+1+j =

and

(S8.2.15)

Et+1 ct+1+j =

j;i ut+i+1 ;

j !

X

i

i=0

+ j+1 ct

j !

X

i

i=0

(S8.2.16)

Et+1 ct+1+j ; Et ct+1+j = j ut+1 :

Substituting in this expression, and noting that ct+1 ; Et ct+1 = ut+1 , we obtain

1

X

j=1

j j ut+1

= 11;;

ut+1 :

Part (ii). For a real consol paying a xed real dividend we have that di;t+1+j = 0,

so the unexpected return is in uenced only by changes in expected future interest rates.

Similar reasoning as in part (i) gives the unexpected real consol bond return as

rb;t+1+j ; Et rb;t+1+j = 1;;

(S8.2.18)

ut+1 :

8.2.3 Part (i). From equation (8.2.7), the equity premium is given by Vce , where

Vce = Covt(ct+1 ; Et ct+1 ; re;t+1 ; Et re;t+1 )

(S8.2.19)

= Cov ut+1 ; 11;;

ut+1

2

= 11;;

u ;

and we may write Cov(; ) instead of Covt(; ) because the process for ct+1 is homoskedastic.

Similarly, the consol bond premium is Vcb , where

2

(S8.2.20)

Vcb = 1;;

u :

(S8.2.17)

Part (ii). The bond premium has the opposite sign to because a positive implies

that a positive endowment shock increases future consumption more than current consumption, so real interest rates rise and bond prices fall when consumption rises. Real

SOLUTION 8.3

41

bonds thus provide a hedge against endowment risk and they have a negative premium.

The bond premium is proportional to the square of
because a larger
both increases

the variability of real interest rates and bond returns, and increases the premium required

by investors for bearing a unit of risk.

Part (iii). The premium of equity over the consol is

u2 ;

(Vce ; Vbe ) = (1

(S8.2.21)

; )

so the equity premium is just the bond premium plus a premium related to dividend

uncertainty, which is always positive and proportional to
.

Part (iv). The lesson for the equity premium literature is that models with high

degrees of risk aversion tend to imply a high bond premium as well as a high equity

premium. This is a counterfactual implication.

Solution 8.3

This can be written as

8

1

< m;

2

(S8.3.1)

w=:

(1 ; a)m; 12

where w is both the individual and aggregate second period endowment.

Consider buying of the asset. The asset price is paid in the second period, so the

expected utility cost is

1 U 0 (m) p + 1 U 0 ((1 ; a) m) p = 1 p 1 +

1

(S8.3.2)

2

2

2 m (1 ; a) m

= 21 p (12;;aa) m :

8.3.1

1

2.

1 U 0 (m) m + 1 U 0 ((1 ; a) m) (1 ; a) m = 1 [1 + 1]

(S8.3.3)

2

2

2

(S8.3.4)

= :

In equilibrium, the expected utility cost must equal the expected utility benet, so

1 p 2 ; a = ;

(S8.3.5)

2 (1 ; a) m

which implies

; a)

(S8.3.6)

p = 2 (1

2 ; a m:

The expected gross return on the claim is the ratio between its expected payo and

its price:

1 m + 1 (1 ; a) m

a2 ;

2

(S8.3.7)

1 + R(a) = 2

=

1

+

p

4 (1 ; a)

which rises with a (a measure of aggregate risk) as we would expect.

8.3.2 Now we have that the individual second period endowment is:

8

1

< m;

2

I

(S8.3.8)

w = : m;

(1

;1 ; a m; b 1 ; b) 12

b

2

42

PROBLEMS IN CHAPTER 8

8

1

< m;

2

A

w =:

(S8.3.9)

(1 ; a)m; 12

Note that we must have b > a so the individual endowment is always non-negative and

log utility is dened. If b = 1 then we are back to the previous case.

Since all agents have the same utility function and face the same probability of being

in each group, they all have the same expected endowment and are identical ex-ante.

However, ex-post their endowments will dier,so there will be ex-post heterogeneity.

As before, consider buying of the asset. The expected utility cost is

1 U 0 (m) p + 1 (1 ; b) U 0 (m) p + 1 bU 0 1 ; a m p

(S8.3.10)

2

b

2

1 p2

1

p

b

a

= 2 m 1 + (1 ; b) + 1 ; a = 2 m 2 + 1 ; a :

b

b

The expected utility gain is

(S8.3.11) 12 U 0 (m) m + 12 (1 ; b) U 0 (m) (1 ; a) m + 12 bU 0 1 ; ab m (1 ; a) m

; a) :

= 12 1 + (1 ; b) (1 ; a) + b (1

1 ; ab

In equilibrium, expected cost equals expected gain so

1

1

p

a

b

(1

;

a

)

(S8.3.12)

2 m 2 + 1 ; ab = 2 1 + (1 ; b) (1 ; a) + 1 ; ab :

Thus

2 (b ; a) + a2 (1 ; b)

(S8.3.13)

m;

p=

2 (b ; a) + ba

The expected gross return on the claim is :

1 m + 1 (1 ; a) m

2

1 + R(b) = 2

(S8.3.14)

p

(2

;

a

)

[2

) + ba] ;

(S8.3.15)

= 4 (b ; a) +(b2;a2a(1

; b)

2

(S8.3.16)

1 + R(b) = 4(2(1;;a)a) = 1 + R(a) ;

and when b = a,

1 + R(b) = 2(2(1;;aa)) = (2 ;2 a) (1 + R(a)):

(S8.3.17)

Since 0 < a 1, R(b) R(a). Therefore, heterogeneity in the form of individual

uninsurable risk increases the expected return on the asset.

8.3.3 The literature on representative agent models tends to nd that average stock returns are higher than can be explained with plausible degrees of risk aversion. Uninsurable

individual risk might be one explanation.

This problem is based on N. Gregory Mankiw's \The Equity Premium and the

Concentration of Aggregate Shocks", Journal of Financial Economics, September

1986.

Mankiw shows that one gets similar results for any utility function with U 000 > 0. Quadratic utility has U 000 = 0 (\certainty equivalence") and uninsurable individual risk has

no eect.

SOLUTION 8.3

43

The result also depends on the fact that there is more dispersion of individual endowments in bad times than in good times.

44

PROBLEMS IN CHAPTER 8

Problems in Chapter 9

Solution 9.1

Without loss of generality, let us consider the random variable pn (T ) (the derivation for

pn (t) is analogous). Denote the moment-generating functions of the increments k in

(9.1.1) and pn (T ) in (9.1.2) as M ( ) and Mp ( ) respectively, where

(S9.1.1)

M ( ) = E[ek ]

(S9.1.2)

= pe + qe;

Pn

(S9.1.3)

Mp ( ) = E[epn (T) ] = E[e k=1 k ]

(S9.1.4)

= E[

(S9.1.5)

n

Y

ek ] =

k=1

e

n

Y

k=1

E[ek ]

+ (1 ; )e;

n

p

p

Mp ( ) = 21 (1 + h )e h +

p

1 (1 ; h )e;ph n

(S9.1.6)

2

p

p n

p

(S9.1.7)

= cosh( h) + h sinh( h)

where cosh(x) and sinh(x) are the hyperbolic sine and cosine functions

x ;x

2

4

cosh(x) = e +2 e

= 1 + x2! + x4! +

3

5

x ;x

= x + x3! + x5! +

sinh(x) = e ;2 e

Simplifying and letting n ! 1 yields

n

2 2

Mp ( ) = 1 + ( 2 + ) Tn + o( n1 )

(S9.1.8)

2 2

(S9.1.9)

! e(+ 2 )T

which is the moment-generating function for a normal random variable with mean T and

variance 2 T .

Denote by [

(S9.2.1)

2 ]0

Solution 9.2

1 @2 L() T

n2

I () = nlim

;

E

!1

0

n @ @ 0 =

45

1

24

46

PROBLEMS IN CHAPTER 9

I ;1 () =

(S9.2.2)

n2

T

0

2 4

Solution 9.3

Consider n observations in the interval [0; T ] equally spaced at intervals h T=n, and let

p(0) = 0 to simplify the algebra. Let pk p(kh). Using (9.3.48) we nd that

(S9.3.1)

Z kh

kh;h

e; (kh;s) dB (s) :

(S9.3.2)

where k N (0; 1) and

(S9.3.3)

Z kh

2 = Var

2

e;
(kh;s) dB (s) =
(1 ; e;2
h ) :

kh;h

We now derive the maximum likelihood estimators ^, ^ and ^2 from which we can obtain

^ and ^ 2 as:

2

^ = ; h1 log(^); ^ 2 = (1 ;2^
e^;2^
h )

(S9.3.4)

by the Principle of Invariance (see Zehna [1966]). The log-likelihood function is given by

(S9.3.5)

n

X

k=1

and the necessary rst-order conditions for the maximum of the log-likelihood function

are

n

@L = 1 X

@

2 k=1[pk ; pk;1 ; h(k ; (k ; 1))]h(k ; (k ; 1)) = 0;

n

@L = 1 X

2

@

k=1[pk ; pk;1 ; h(k ; (k ; 1))](pk;1 ; h(k ; 1)) = 0;

n

@L = ; n 1 + 1 X

2

@2

2 2 24 k=1[pk ; pk;1 ; h(k ; (k ; 1))] = 0:

(S9.3.6)

(S9.3.7)

(S9.3.8)

k=1 Pn

;

[k ; ^(k ; 1)]2

Pn (pkk=1

k=1Pn ; ^hk)(pk;1 ; ^h(k ; 1)) ;

^ =

k=1 [pk;1 ; ^h(k ; 1)]2

n

1X

2

2

^ =

k=1

SOLUTION 9.3

47

From here we will assume that the trend is known exactly. Then we can calculate

n

@2L = ; 1 X

2

@2

2 k=1[pk;1 ; h(k ; 1)] ;

@2L = ; 1 @L ;

@@2

2 @

n

2

@L = n1 ; 1X

2

@ (2 )2

2 4 6 k=1[pk ; pk;1 ; h(k ; (k ; 1))] :

But observe that

" 2 #

P

limn!1 n1 nk=1 (pk;1 ; h(k ; 1))2

1 @ L

lim

;

E

=

n!1

n @2 (^;^2 )

limn!1 ^2

= 1 ; e1;2
h ;

" 2 #

1 @L

lim

;

E

n!1

n @@2 (^;^2 ) = 0;

" 2 #

1 n 1 1

1 @L

lim

;

E

=

lim

;

E

n!1

n!1

n @ (2 )2 (^;^2 )

n (; 2 ^4 ) = 24 :

Using (9.3.7), we conclude that

pn(^ ; ) a N (0; 1 ; e;2
h );

(S9.3.9)

p

(S9.3.10)

n(^2 ; 2 ) a N (0; 24 ):

The asymptotic distribution of
and 2 can now be obtained using (S9.3.4) and the delta

method described in the Appendix A.4:

pn(^
;
) a N 0; 1 (e2
h ; 1) ;

(S9.3.11)

h2

e4
h

p

a

2

2

4

;

2
h

2

(S9.3.12) n(^ ; ) N 0; 2 1 + 2
2 h2 [1 ; e (1 + 2
h)] :

To derive the continuous-record asymptotics of
^ and ^2 , we let n ! 1 while T is

held xed, hence h = T=n ! 0. Since

(S9.3.13)

^ = ; h1 log(1 ; (1 ; ^)) = h1 (1 ; ^) + o(h) = (1 ;1 ^) + o( n1 );

n

we conclude that

Pn

; pk;1 ; n1 )(pk;1 ; k;n 1 ) :

^ a ; k=1 (p1k P

(S9.3.14)

n

k ;1 2

n k=1 [pk;1 ; n ]

The denominator converges to

ZT

(S9.3.15)

while the numerator converges to

(S9.3.16)

We conclude that

(S9.3.17)

ZT

0

^ a ;

ZT

0

0

:

R T (p(s) ; s)02 ds

0

48

which simplies to

PROBLEMS IN CHAPTER 9

T q (s) dq (s)

^ a ; R0 T 2 :

0 q (s) ds

where q(t) p(t) ; t. Finally, it can be shown that

n

X

^2 = n1 [qk ; ^qk;1 ]2 a 2 :

(S9.3.19)

k=1

(S9.3.18)

Solution 9.4

9.4.1 The maximum likelihood estimates (9.3.27), (9.3.28) are evaluated using daily returns from January 2, 1991 to December 29, 1995 and assuming h = 1=253, i.e., 253

trading periods in a year. The riskfree interest rate is set to r = 5%. The estimates are:

2

^ 2 = 0:074 ; ^ r ; ^2 = 0:0131 :

9.4.2 Two variants of the Monte Carlo method are used:

1. The crude method of Section 9.4.1.

2. The antithetic variates method of Section 9.4.4.

The initial stock price (the closing price on December 29, 1995) is $91.375. 100; 000

replications are used in both cases (m = 100; 000). The number of discrete intervals is

n = 253.

The crude Monte Carlo method produces an estimate according to (9.4.6) of:

H^ (0) = $17:70 :

The standard deviation of the estimate H^ (0) is estimated according to (9.4.10) as:

^y (253) = $19:60 :

Therefore, according to (9.4.8), a 95% condence interval is

$17:58 H (0) $17:82 :

The minimum number of replications necessary to yield a price estimate within $0:05 of

the true price is estimated according to (9.4.9):

m 5:905 105 :

The antithetic variates method produces an estimate according to (9.4.13):

H^ (0) = $17:63 :

Standard deviation of the estimate H^ (0) is estimated according to (9.4.15):

^y (253) = $8:56 :

As a result, according to (9.4.8), a 95% condence interval is

$17:58 H (0) $17:69 :

The minimum number of replications necessary to yield a price estimate within $0:05 of

the true price is estimated according to (9.4.9) as:

m 1:126 105 :

9.4.3 The closed-form solution for the option price is given by the Goldman-Sosin-Gatto

formula (9.4.11) and is evaluated using the estimate of 2 obtained in Problem 9.4.1:

H (0) = $18:91 :

The dierence between the theoretical price H (0) and our estimate H^ (0) arises from

the dierence between the maximum of discretely-sampled and continuously-sampled

SOLUTION 9.4

49

prices. Specically, the theoretical price H (0) of the option is evaluated under the assumption that the option allows one to sell the stock at the maximum price observed over

the course of the entire year. The estimate H^ (0) was obtained under the assumption that

only daily closing prices are used to evaluate the maximum. Obviously, the rst denition

always leads to a higher option price than the second.

In the context of this particular problem the second denition of the option (the one

used in Monte Carlo simulations) is more relevant, since it is based on the denition of

the actual option. The Goldman-Sosin-Gatto formula is a continuous-time approximation

to this option. Therefore, the Monte Carlo estimator of the option price should be used

to decide whether to accept or reject CLM's proposal.

50

PROBLEMS IN CHAPTER 9

Problems in Chapter 10

Solution 10.1

0:4868 per dollar of their face values. Since nominal interest rates cannot be negative, the

nding that PA < PB implies an arbitrage opportunity and is inconsistent with any expectations theory.

0

;70:091 0:5289 and PB0 = e;80:08

10.1.2 Prices of zero-coupon bonds are now PA = e

0:5273 per dollar of their face values. As PA0 PB0 in this case, the prices do not imply

an arbitrage opportunity and may be consistent with the pure expectations hypothesis.

10.1.3 Let us assume the coupon payments are annual and are made at the end of the

year. Consider rst the case analogous to Problem 10.1.1. Prices do not now imply an

arbitrage opportunity. As an example, assume that all one- to eight-year zero-coupon

bonds have price P8 per one dollar of their face value, and that the nine-year zero-coupon

bond has price P9 . Under these assumptions we can express the prices as

(S10.1.1)

P8 = 1 + 8PA 0:08 0:2944;

8 0:08P8 0:2752:

(S10.1.2)

P9 = PB ;1 +

0:08

10.1.1

We see that, under this non-stochastic term structure given by P8 and P9 , all interest

rates are nonnegative and P8 P9 , so that no arbitrage opportunity exists.

Now, consider the case analogous to Problem 10.1.2. Assume that all one- to sevenyear zero-coupon bonds have price P70 per one dollar of their face value and that eight-year

zero-coupon bond has price P80 . Under these assumptions we can express the prices as

0

P70 = 1 + 7PA 0:08 0:3390;

(S10.1.3)

0 7 0:08P70

(S10.1.4)

0:3125:

P80 = PB ;1 +

0:08

Note however that the assumptions required to rationalize these bond prices are rather

extreme, since they require zero nominal interest rates between one and eight years. The

loglinear approximate model for coupon bonds presented in (10.1.20) gives a dierent

answer. This model eectively imposes \smoothness" on the term structure. Equation

(10.1.20) allows us to compute the implicit n-period-ahead 1-period log forward rate given

the coupon-bond duration Dcnt in (10.1.10), which in turn requires the coupon-bond price

Pcnt in (10.1.9).

For the data in Problem 10.1.1 we have

(S10.1.5)

Pc8t = :9171; Dc9t = 6:1186 years

Pc9t = :9797; Dc9t = 6:7212 years;

so (10.1.20) gives

f8t ;3:1684% < 0

51

52

PROBLEMS IN CHAPTER 10

Notice that the bonds are not selling at par, so it is not correct to use the simpler formula

for Dcnt that obtains in this case. This result again implies under the log pure expectations

hypothesis a negative one-period log yield 8 periods ahead.

Similarly, for the data in Problem 10.1.2 we have

(S10.1.6)

Pc7t = :9245; Dc7t = 5:5615 years

Pc8t = :9813; Dc8t = 6:1876 years;

so (10.1.20) gives

f7t ;1:7711% < 0

which implies under the log pure expectations hypothesis a negative one-period log yield

7 periods ahead. Using this approach, we nd that coupon bonds violate the log PEH

more than zero-coupon bonds. The reason is that the duration of coupon bonds does

not increase linearly with their maturity, but increases at a decreasing rate. That is,

Dc;n+1;t ; Dc;n;t < 1. This in turn makes it easier to get negative forward rates for given

yields.

Solution 10.2

Assume the postulated process and simplify notation, introducing at yt1 ; y1;t;1

and bt y2t ; y1t . The equations of the model can then be written as

I

at = bt + t ;

II

bt = 21 Et [at+1 ] + xt;

(S10.2.1)

III

xt = xt;1 + t ;

IV

at =
xt + t :

>From the rst and fourth equations we get bt =
;1 xt; from the third and fourth

equations we get Et [at+1 ] =
xt; the second equation then gives an expression for the

coecient
in terms of the other parameters of the model,

= 2 ;2 :

(S10.2.2)

It is straightforward to verify that with this value for
, the y1t process satises all the

equations of the model, provided that < 2.

10.2.2 Using notation from Problem 10.2.1, the regression has the form

(S10.2.3)

at+1=2 = + bt + ut+1 :

As Et [at+1 ] =
xt and bt =
;1 xt, we see that the population parameters are = 0

and = =2. Clearly < 1 since we have required < 2.

10.2.3 Assume the process of the given form and simplify notation, introducing at

y1t ; y1;t;1 and bt ynt ; y1t . Note that

(S10.2.4)

yn;t+1 ; ynt = bt+1 + at+1 ; bt:

The equations of the model and of the postulated process are then

I

at = bt + t ;

(S10.2.5)

II

bt = (n ; 1)Et [bt+1 + at+1 ; bt] + xt;

II

xt = xt;1 + t ;

IV

at =
xt + t :

>From the rst and fourth equations we get bt =
;1 xt ; from the third we get Et [bt+1 ] =

;1 xt; from the third and fourth we get Et [at+1 ] =
xt; and the second equation then

10.2.1

SOLUTION 10.2

53

= n ; (n ; 1)(1 + ) :

(S10.2.6)

It is straightforward to verify that with this value of , the y1t process satises all the

equations of the model, provided that (1 + )(n ; 1) < n.

In our notation, the regression takes the form

bt+1 + at+1 ; bt = + n b;t 1 + ut+1 :

(S10.2.7)

As E[bt+1 +at+1 ;bt] = ( ;1 + ; ;1 )xt and bt = ;1 xt, we see that the population

parameters are = 0 and = (1 + )(n ; 1) ; (n ; 1). The parameter restrictions we

have imposed allow to be either positive or negative.

10.2.4 The model does explain why short-rate regressions of the type explored in Problem

10.2.2 give coecients positive but less than one, while long-rate regressions of the type

explored in Problem 10.2.3 often give negative coecients. The underlying mechanism is

a time-varying term premium, interacting with the desire of the monetary authority to

smooth interest rates.

A limitation of this model is that it assumes a nonstationary interest rate process,

which has unsatisfactory long-run properties. For example, with probability one the interest rate eventually becomes negative. Bennett McCallum, \Monetary Policy and the

Term Structure of Interest Rates", NBER Working Paper No. 4938, 1994, works out a

stationary version of this model; the algebra is more complicated but the properties of the

model are similar.

54

PROBLEMS IN CHAPTER 10

Problems in Chapter 11

Solution 11.1

We assume throughout the problem that bond prices are determined by the homoskedastic lognormal model implied by equations (11.1.5) and (11.1.3),

(S11.1.1)

;mt+1 = xt + t+1

(S11.1.2)

xt+1 = (1 ; ) + xt + t+1;

2

with t N (0; ), but to t the current term structure of interest rates we assume instead

that the state variable follows the process given in equation (11.3.4):

(S11.1.3)

xt+i = xt+i;1 + gt+i + t+i :

A useful way to relate the deterministic drift terms gt+i and the parameters of the

true pricing model when tting the term structure of interest rates is to compute the

forward rates implied by the assumed model (S11.1.1) and (S11.1.3), and compare them

with those implied by the true model (S11.1.1) and (S11.1.2). To compute the forward

rates implied by the assumed model we need rst to compute the log bond prices, since

fn;t = pn;t ; pn+1;t . Using equality (11.0.2) and the lognormal property of the stochastic

discount factor, we have that

11.1.1

"Y

n

pn;t = log Et

"X

n

i=1

Mt+i

n

X

mt+i :

= Et

mt+i + 21 Vart

i=1

i=1

But from the assumed model for the state variable (S11.1.3) we have

xt+i = xt +

(S11.1.4)

so

n

X

i=1

mt+i = ;

n

X

i=1

(S11.1.5)

pn;t = ;n xt ;

j=1

gt+j +

xt+i;1 ;

= ;n xt ;

and

Xi

n

X

i=1

n

X

i=1

n

X

i=1

Xi

j=1

t+j ;

t+i

(n ; i) gt+i ;

n

X

i=1

( + n ; i)2 t+i

n

X

(n ; i) gt+i + 12 ( + n ; i) 2 :

i=1

We can now use (S11.1.5) to compute forward rates implied by the assumed model:

fn;t = pn;t ; pn+1;t

n

X

= xt + gt+i ; 21 ( + n)2 2 :

(S11.1.6)

i=1

55

56

PROBLEMS IN CHAPTER 11

Comparing (S11.1.6) with equation (11.1.14), that gives us the forward rates implied by

the true model, we nd immediately that the drift terms gt+i are related to the parameters

of the true model by the following expression:

"

#

n

n 2

X

1

1

;

n

2

gt+i = ; (1 ; ) (xt ; ) ; 2 + 1 ; ; ( + n) 2 :

(S11.1.7)

i=1

11.1.2 Since r1;t+1 = ;Et [mt+1 ] ; Vart (mt+1 ) =2, the short term interest rates at (t + 1)

implied by the assumed model and the true model are the same:

(S11.1.8)

r1;t+1 = xt ; 12 2 2 :

The dynamics of the state variable in the true model, given by (S11.1.2), and (S11.1.8)

imply that future short rates equal:

n

X

r1;t+n+1 = (1 ; n ) + n xt + n;i t+i ; 21 2 2

i=1

= r1;t+1 ; (1 ; n ) (xt ; ) +

n

X

i=1

n;i t+i;

so the expected future log short rates in the true model are

(S11.1.9)

Et [r1;t+n+1 ] = r1;t+1 ; (1 ; n ) (xt ; ) :

The dynamics of the state variable in the assumed model, given by (S11.1.3), imply:

n

n

X

X

r1;t+n+1 = xt + gt+i + t+i ; 12 2 2

i=1

i=1

= r1;t+1 +

n

X

i=1

gt+i +

n

X

i=1

t+i;

so expected future log short rates under the assumed model are

(S11.1.10)

Et [r1;t+n+1 ] = r1;t+1 +

(S11.1.11)

n

X

i=1

n

X

i=1

gt+i:

gt+i = ; (1 ; n ) (xt ; ) ;

the assumed model will be able to reproduce the expected short rates. However, by

comparing (S11.1.7) and (S11.1.11) we can see that it is not possible to choose drift terms

so they match simultaneously both current forward rates and expected future log short

rates, since

1 ; n 2

+ 1 ; 6= ( + n)2

unless ! 1, i.e., unless the state variable in the true model follows a random walk. It is

also interesting to note that the set of deterministic drifts that matches expected future

log short rates|see equation (S11.1.11)|converges to ;(xt ; ) as n ! 1, while the set

of deterministic drifts that matches forward rates|see equation (S11.1.7)| tends to ;1

as n ! 1. Therefore, if we choose the drift terms so they reproduce the forward rate

structure of the true model, this will result in expected future log short rates declining

without bound as we increase the horizon, while the true model implies that the expected

future log short rates converge to a nite constant.

SOLUTION 11.1

57

>From equation (11.1.8) and (S11.1.2), the time t conditional variance of log bond

prices at time t + 1 implied by the true bond pricing model is

11.1.3

n 2

= 11;; 2 ;

(S11.1.12)

while from (S11.1.5) and (S11.1.3), the time t conditional variance of log bond prices at

time t + 1 implied by the assumed bond pricing model is

(S11.1.13)

Vart (pn;t+1 ) = n2 2 :

Hence (S11.1.13) cannot be equal to (S11.1.12) unless ! 1, i.e. unless the state variable

follows a random walk in the true model. Moreover, for n > 1, the conditional variance

of log bond prices implied by the assumed model is larger than the conditional variance

implied by the true model and, while the true model implies that the conditional variance

of log bond prices is bounded at 2 =(1 ; )2 as n ! 1, the assumed model implies an

unbounded conditional variance.

11.1.4 Section 11.3.3 shows that the price of a European call option written on a zerocoupon that matures n + periods from now, with n periods to expiration and strike price

X , is given under the true model by

where Pn;t = expfpn;t g = expfAn + Bn xtg is the price of the bond, () denotes the

cumulative distribution function of a standard normal random variable,

d1 = pn+;t ; xp; pn;t + Vart (p;t+n) =2 ;

p Vart (p;t+n)

d2 = d1 ; Vart (p;t+n);

x = log(X ) and

Vart (p;t+n) = B2 Vart (xt+n)

2 1 ; 2n

2

(S11.1.14)

= 11;;

1 ; 2 :

In our assumed model we use the same formula to value the option, except that we

need to compute Vart (p;t+n) under our assumed process for the state variable (S11.1.3).

From (S11.1.5), we have

(S11.1.15)

= 2 n2 ;

Obviously, (S11.1.15) diers from (S11.1.14), unless ! 1, so in general the assumed

model will misprice options. For > 1 and/or n > 1, it will overstate the volatility of

the future log bond price, hence overvaluing the option. This overvaluation increases with

the expiration date of the option and/or the maturity of the underlying bond. This is

true no matter what combination of the drift parameters we choose. Backus, Foresi and

Zin (1996) use this result to caution against the popular practice among practitioners of

augmenting standard arbitrage-free bond pricing models with time-dependent parameters

to t exactly the yield curve. This augmentation may seriously misprice state-contingent

claims, even though it is able to exactly reproduce the prices of some derivative securities.

58

PROBLEMS IN CHAPTER 11

Solution 11.2

(S11.2.1)

;mt+1 = xt + t+1

11.2.1

(S11.2.2)

xt+1 = (1 ; ) + xt + t+1:

Thus, the price function for an n-period bond is

(S11.2.3)

;pnt = An + Bn xt

with

n

Bn = 1 + Bn;1 = 11;; ;

(S11.2.4)

(S11.2.5)

An ; An;1 = (1 ; )Bn;1 ; ( + Bn;1 )2 2 =2;

and A0 = B0 = 0.

Equation (11.3.15) in CLM gives the price at time t of an n-period forward contract

on a zero coupon-bond which matures at time t + n + as Gnt = P+n;t =Pnt . Taking

logs, gnt = p+n;t ; pnt. Substituting out p+n;t and pnt using (S11.2.3)-(S11.2.5) yields:

(S11.2.6)

;gnt = (An+ ; An ) + (Bn+ ; Bn )xt:

Thus, the pricing function for an n-period forward contract on a zero coupon-bond

which matures at time t + n + is given by:

g xt ;

(S11.2.7)

;gnt = Agn + Bn

with

Agn = An+ ; An

g = Bn+ ; Bn ;

Bn

where (S11.2.4) and (S11.2.5) can be used to write An+ , Bn+ as functions of An , Bn .

Clearly, the log forward price gnt is ane in the state variable xt.

In order to show that the log futures price hnt is also ane in the state variable we

can use equation (11.3.10) in CLM:

(S11.2.8)

Hnt = Et [Mt+1 H;n;1;t+1=P1t ]

Taking logs and assuming joint lognormality:

(S11.2.9) hnt = Et [mt+1 + h;n;1;t+1 ; p1t ] + 21 Vart [mt+1 + h;n;1;t+1 ; p1t ] :

Let us rst determine h1t . Since h;0;t+1 = p;t+1we have that:

(S11.2.10)

h1t = Et [mt+1 + p;t+1 ; p1t] + 12 Vart [mt+1 + p;t+1 ; p1t ]

Substituting out mt+1 using (S11.2.1) and p;t+1, p1t using (S11.2.3) and (S11.2.2) yields:

(S11.2.11)

B xt ; B t+1 + xt ; 2 2 =2 +

1 Var ;x ; ; A ; B (1 ; );

t+1

2 t t

B xt ; B t+1 + xt ; 2 2 =2 :

Since Et t+1 = 0 and Vart t+1 = 2 it follows that:

h xt

(S11.2.12)

;h1t = Ah1 + B1

SOLUTION 11.2

with

59

Ah1 = A + (1 ; )B ; 2 =2 ; 2 + ( + B )2

h = B :

B1

h xt . We proceed to verify

Let us now solve for hnt . We guess that ;hnt = Ahn + Bn

h as functions

our guess. At the same time we derive formulas for the coecients Ahn , Bn

of the term structure coecients An , Bn . Proceeding as above:

(S11.2.13) hnt = Et [mt+1 + h;n;1;t+1 ; p1t] + 21 Vart [mt+1 + h;n;1;t+1 ; p1t ]

and using our guess to substitute out for h;n;1;t+1 :

h (1 ; );

hnt = Et ;xt ; t+1 ; Ah;n;1 ; B;n

;1

(S11.2.14)

h xt ; B h t+1 + xt ; 2 2 =2

B;n

;1

;n;1

1 Var ;x ; ; Ah ; B h (1 ; );

t+1

;n;1

;n;1

2 t t

h xt ; B h t+1 + xt ; 2 2 =2 :

B;n

;1

;n;1

We obtain:

(S11.2.15)

h

i

h (1 ; ) ; 2 =2 ; 2 + ( + B h )2 + B h xt :

;hnt = Ah;n;1 + B;n

;1

;n;1

;n;1

Thus

h

i

h (1 ; ) ; 2 =2 ; 2 + ( + B h )2 ;

Ahn = Ah;n;1 + B;n

;1

;n;1

h = B h :

Bn

;n;1

h

Solving recursively and using B1 = B yields

h (1 ; ) ; 2 =2 ; 2 + ( + n;1 B )2 ;

(S11.2.16)

Ahn ; Ah;n;1 = B;n

;1

h = n B :

Bn

11.2.2 The log ratio of forward to futures prices is given by

h xt :

(S11.2.17)

gnt ; hnt = (An ; An+ ) + (Bn ; Bn+ )xt + Ahn + Bn

h = 0.

In order to show that this is constant we need to show that Bn ; Bn+ + Bn

Straightforward algebra gives us:

h = Bn ; Bn+ + n B =

(S11.2.18) Bn ; Bn+ + Bn

n

n+

n

n+

= 1 ; ; 1 + 1 ; + ; = 0:

Showing that the ratio of forward to future prices is greater than one is equivalent to

showing that the log ratio is greater than zero. In order to do so we write

gnt ; hnt = An ; An+ + Ahn =

= An;1 + (1 ; )Bn;1 ; ( + Bn;1 )2 2 =2

;An+ ;1 ; (1 ; )Bn+ ;1 + ( +h Bn+ ;1 )2 2 =2

i

h

2

2

h

2

+Aht;n;1 + (1 ; )B;n

;1 ; =2 ; + ( + B;n;1 ) :

60

PROBLEMS IN CHAPTER 11

It can easily be checked that the terms in (1 ; ) add up to zero. Given the recursive

nature of the problem and remembering that g1t ; h1t = 0 we have that

(S11.2.19) gnt ; hnt = ;2 =2

Using the fact that j B

(S11.2.20)

Thus we have that

(S11.2.21)

nX

;1 h

j=0

( + Bj )2 ; ( + Bj+ )2 ; 2 + ( + j B )2 :

2

; n;1 )(1 ; n )

:

gnt ; hnt = (1 ; (1)(1

; )3 (1 + )

gnt ; hnt < 0 when ; 1 < < 0:

The dierence between a futures contract and a forward contract is that the rst is

marked to market each period during the life of the contract, so that the purchaser of

a futures contract receives the futures price increase or pays the futures price decrease

each period. When interest rates are random, these mark-to-market payments may be

correlated with interest rates. When 0 < < 1, so that Cov(hnt ; y1t ) < 0, the purchaser

of a futures contract tends to receive the futures price increase at times when interest rates

are low, and tends to pay the futures price decrease at times when interest rates are high,

making the futures contract worth less than the forward contract. On the other hand,

when ;1 < < 0, the futures contract will be worth more than the forward contract

since its purchaser tends to receive price increases when interest rates are high (so that

the money can be invested at a high rate of return).

2

2

11.2.3 The parameter values for this part, = 0:98 and = 0:00051 , can be found in

Section 11.2.2, page 453 (and not in Section 11.1.2).

Problems in Chapter 12

Solution 12.1

12.1.1

There are several criteria with which random number generators can be judged:

erties of generated sample and assessed by batteries of statistical tests of independence, goodness-of-t to specic probability distributions, etc.

Computational eciency, in terms of cost of implementation, resource requirements, volume of output per second, volume of output in absolute terms, all without deterioration of stochastic quality.

Portability of the algorithm.

Reproducibility of random series (based on the initial \seed" of the random number

generator).

The ultimate introduction to the science and art of pseudorandom number generation

is Chapter 3 of D. E. Knuth's classics The Art of Computer Programming 1969, 1981, where

the most in
uential and comprehensive study of the subject is to be found.

One example of the many recent treatises on the state of the art is Fishman (1996),

which emphasizes pseudorandom number generators in Chapter 7. High-quality pseudorandom number generators also emerge in cryptography. Cryptographically secure generators, related to stream ciphers and one-way hash functions achieve extraordinary stochastic quality, generally at the expense of increasing computation costs. See for example

Schneier (1996, Chapters 16{18).

There exist batteries of statistical tests intended to measure stochastic quality of pseudorandom number generators; these include tests such as chi-square, Kolmogorov-Smirnov,

frequency, serial, gap, permutation, run, moments, serial correlation, and especially spectral tests (see Knuth [1969] for details); or, for example, an omnibus test assessing joint

independence and one- to three-dimensional uniformity, assembled by Fishman (1996,

Section 7.12).

12.1.2 Generally, very well researched and tested MLCG generators constitute an accepted

pragmatic compromise among the criteria imposed on pseudorandom number generators

discussed in Problem 12.1.1. The proper choice of parameters of MLCG generators is

essential, and theoretical guidelines are readily available in Knuth (1969) and elsewhere.

The quality of the tent- and logistic-map generators is inferior for most purposes, as most

standard statistical tests of randomness will show.

The extra modication of using parameters like 1.99999999 instead of 2 etc. patches

the most obvious
aw of the tent- and logistic-map generators: with real numbers represented in binary form using nite-length mantissas, repetitive multiplication by 2 deteriorates quality of the sequence rapidly, i.e., the sequence degenerates in time that is

proportional to the mantissa length. In most practical cases, though, the use of wellresearched pseudorandom number generators with solid theoretical guarantees of quality,

such as MLCG, is indicated.

If the quality of even properly chosen MLCG is not sucient for an application at

hand, one may consider using some other classes of well-tested generators with balanced

61

62

PROBLEMS IN CHAPTER 12

Table Estimates of kernel-regression betas of IBM relative to S&P 500 based on monthly

return data from 1965:1 to 1994:12. Each estimate is local to a particular level of S&P

500 monthly return.

SP500 [%] ^IBM;SP500

;15

;10

;5

0

5

10

15

1.366

1.395

0.689

0.666

0.806

0.531

1.994

and Zaman [1991]).

Solution 12.2

Equations (12.4.1) and (12.4.3) describe one unit. Our case involves ten such units with

J 5. The output layer is given by equation (12.4.4) with K 10. For simplicity,

choose h() to be the identity, in accord with the discussion on pages 514{542. Thus,

the nonlinear model has 60 parameters to t. Using a nonlinear optimization technique

of choice, nd the parameter values that attain the minimum (beware of local minima!)

in-sample root-mean-squared-error (RMSE) of the one-step-ahead estimate, with identical

weights given to each datapoint of S&P 500 returns from 1926:1 to 1985:12. Then, apply

the tted perceptron parameters on data in period from 1986:1 to 1994:12.

The RMSE will be substantially larger in the out-of sample period than in the insample period. The out-of-sample RMSE 60-parameter perceptron will probably not be

drastically smaller than out-of-sample RMSE of a linear model with less immodest number

of parameters (say, ten (10); consider an OLS regression with ve lagged returns and

their squares as explanatory variables), but the in-sample RMSE of the former will be

noticeably smaller than RMSE of the latter. This phenomenon can be related to concept

of \overtting" which occurs when lack of structural, qualitative information of the data

generating stochastic process is countered by increase in number of ad hoc degrees of

freedom in the model: this procedure results in excellent in-sample t while out-of-sample

performance stays mediocre.

Solution 12.3

First implement the kernel regression estimator m^ h (x) according to formula (12.3.9) with

a Gaussian kernel Kh (x) as in (12.3.10). Second, determine optimal bandwidth by minimizing the cross-validation function CV(h) as in (12.3.13), based on estimator m^ h (x) and

given historical S&P 500 and IBM monthly returns.

Numerically, the appropriate bandwidth for period from 1965:1 to 1994:12 is h =

1:49% (the scale is in monthly returns of S&P 500). The resulting regression is plotted

(Figure 12.1).

The analog of the conventional beta estimate here is the quantity @ m^ h (x)=@x, evaluated at particular level of S&P 500 return x. See Section (12.3.3) for a detailed discussion

of average derivative estimators.

Let us replace derivative by its discrete analog with a step length dierence of 1% of

the S&P 500 monthly return. The resulting estimates of 's for dierent levels of S&P

500 returns is shown in Table 12.1.

SOLUTION 12.3

63

20

0

-10

10

-20

-20

-10

10

We see that the local estimates of beta vary considerably, most likely due to the

relatively small number of datapoints in the estimation, possible variation in beta over

time, or genuine nonlinearity of the relation between IBM and S&P 500 monthly returns.

Some advantages of kernel regression relative to ordinary least squares are: crossvalidation allows for nonparametric, adaptive and asymptotically consistent estimation of

the true relation between IBM and S&P 500 returns even when this relation is not linear;

the kernel estimator m^ (x) conveys more information about the relationship than a single

parameter ( ) and allows easy visualization of the relation.

- Guide Nut 2007 DRAFT 2Uploaded bySunita Abraham
- THT 3Uploaded byDan BirdHunter Fallins
- Lec08Uploaded bydavid
- Why Does Higher Q Make Better Clocks_vs2Uploaded byJohnHaine
- Imaging Process StepsUploaded byErik Martinez
- PPT-7Uploaded byzeeshan
- BootStrapping.pdfUploaded byManish Kumar
- Statistik Uji Box MUploaded byAyunning Tieas
- Standard Deviation FormulasUploaded byYoussef Khairane
- Chapter 4Uploaded byNorazlina Mohd Sabar
- norway04_nonparametricUploaded byshree_saha
- Rci-10 Scoring and NormsUploaded bylavinia23
- MacCallum Et Al 1999 Sample Size in FAUploaded bydrdjettel
- third revisionUploaded byapi-3759646
- Nie BoerUploaded byForce Mapu
- 36_1Uploaded byMahyar Zobeidi
- 82011987.pdfUploaded byHelen Yong
- 2-1 a Thomas Homogeneity ISO 13528 TemplateUploaded byAgus Kurniawan
- Factor Model Risk Lecture HandoutUploaded byronnie1987
- Descriptive StatisticsUploaded byGizem Ünver
- Prime - Cpm - Unit 5Uploaded bysalman
- Horwitz 2012Uploaded byFebby Abie Rakha
- SolutionUploaded bySivaMaroju
- Effective Sampling Method for Estimating Bending Strength DiUploaded byChandra Clark
- ASTM E 1490 – 03Uploaded byFrancisco Guerra
- 917-2943-2-PBUploaded bykofi mayhew
- Airflow Quantity MeasurementsUploaded byadam fakhri
- Dof Anamet35 (Notes)Uploaded byJamall Clt
- A Survey on TOA Based Wireless Localization and NLOS Mitigation TechniquesUploaded byazizbahmani
- 9781447155706-c2Uploaded byFabrizio Marocco

- Big Data Analytics with Storm, Spark and GraphLabUploaded byImpetus
- Engineering Mechanics Statics 6th Edition Meriam Kraige Solutions ManualUploaded bythehighlife1080
- One Act PlayUploaded byjstreet
- Statement on the Amuru-Adjumani Boundary Land ConflictUploaded byAfrican Centre for Media Excellence
- google glassUploaded bybill power
- LTE Event A3.docxUploaded byeduardo_pl8256
- Wheatley MalesUploaded byFeli Popescu
- Craig Metaethical Foundations For MoralityUploaded byyellowforkloyalty
- Low Molecular Weight HeparinUploaded byFrancisco Javier Calvo
- shivanjalisinghresume 2017Uploaded byapi-354527690
- Mid ManagementUploaded byAjanudin Montilalu
- Dcs Assignment No 2Uploaded byramjee26
- Steady-Flow Energy Equation - MITUploaded byJúlio Gabriel Queiroz dos Santos
- Even and Odd Math Proofs SolutionsUploaded byalberthawking
- Pre Divorce Private Coaching QuestionnaireUploaded byKassandra Vaughn
- Report on racial profiling, discrimination by Toronto police toward the black communityUploaded byCityNewsToronto
- Interaction ModelingUploaded byabcyoyoyo12
- Mass Customization and FootwearUploaded byNelson Oliveira
- Urban Governance of Employment ActivationUploaded byPostmisfit
- PROCEDURAL VERSUS SUBSTANTIVE JUSTICE: RAWLS AND NOZICKUploaded bySergio Lobos
- penciuman 10 3Uploaded byviana kusuma
- Deming CriteriaUploaded byckbiswas
- Safety Management System EMA Report TUVUploaded bysaadullah
- National Conference on Health Communication, Marketing, and Media.Twitter Transcript (#HCMMconf)Uploaded byAnna Taylor- Ellis
- PRAVEEN REPORT (1).docxUploaded byabdul imran
- Domnich.2011.JACerS.pdfUploaded bymateng011
- Military Handbook Mil Hdbk 415aUploaded bythomas.bhatia
- Www.pythian.com News 654 Installing Oracle 11g on Ubuntu Linux 710 Gutsy GibbonUploaded bychhari45
- BC0058.docxUploaded byvinodnirala
- 110921_Commissioning of Offshore Installations in Building YardsUploaded byJeff Zhang

## Much more than documents.

Discover everything Scribd has to offer, including books and audiobooks from major publishers.

Cancel anytime.